The Economics of "Radiator Springs": Industry Dynamics, Sunk Costs, and Spatial Demand Shifts

A Service of

Leibniz-Informationszentrum econstor Wirtschaft Leibniz Information Centre Make Your Publications Visible. zbw for Economics

Campbell, Jeffrey R.; Hubbard, Thomas N.

Working Paper The economics of "radiator springs": Industry dynamics, sunk costs, and spatial demand shifts

Working Paper, No. 2009-24

Provided in Cooperation with: Federal Reserve Bank of Chicago

Suggested Citation: Campbell, Jeffrey R.; Hubbard, Thomas N. (2009) : The economics of "radiator springs": Industry dynamics, sunk costs, and spatial demand shifts, Working Paper, No. 2009-24, Federal Reserve Bank of Chicago, Chicago, IL

This Version is available at: http://hdl.handle.net/10419/70526

Standard-Nutzungsbedingungen: Terms of use:

Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.

Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise use the documents in public. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated licence. www.econstor.eu

The Economics of “Radiator Springs:” Industry Dynamics, Sunk Costs, and Spatial Demand Shifts

Jeffrey R. Campbell and Thomas N. Hubbard

Federal Reserve Bank of Chicago WP 2009-24

The Economics of “Radiator Springs:” Industry Dynamics, Sunk Costs, and Spatial Demand Shifts

Je¤rey R. Campbellyand Thomas N. Hubbardz November 30, 2009

Abstract We measure industry evolution following permanent changes in the level and location of demand for gasoline in hundreds of counties during the time surrounding the completion of Interstate Highway segments. We …nd that the timing and margin of adjustment depends on whether the new highway is located close to or far from the old route. When the new highway is close to the old one, there is no evidence that the number of stations changes around the time it opens. However, average station size increases by 6% before the highway is completed. When the new highway is far from the old one (say, 5-10 miles), the number of stations increases by 8% and average station size remains unchanged. Unlike the station size adjustment when the new highway is close, the entire increase takes place after construction. These results provide evidence on how this industry, which is characterized by high location-speci…c sunk costs, adjusts to demand changes. Our results are consistent with theories in which …rms have strategic investment incentives to preempt competitors

We thank Steven Berry for his thoughtful discussion of an earlier version of this paper. Chris Ody provided exceptional research assistance. The views expressed herein are those of the authors. They do not necessarily re‡ect the views of the Federal Reserve Bank of Chicago, the Federal Reserve System, or its Board of Governors. yFederal Reserve Bank of Chicago zNorthwestern University

1 1 Introduction

The construction of the Interstate Highway System in the United States and its impact on local industries continues to be part of American folklore and popular culture, over 50 years after construction started and over 25 years after it was essentially completed. The plot of Disney/Pixar’s2005 animated feature "Cars" concerns a town ("Radiator Springs") whose population had severely declined after it was bypassed by an Interstate highway. Radiator Springs is nearly empty and devoid of through tra¢ c, and had been so for years. However, several local businesses had yet to exit the market, including a gas station and an auto repair shop. For economists interested in industry dynamics, the construction of In- terstate Highways o¤ers an interesting opportunity: completion of highways is observable, and represents permanent demand shifts for highway-related services. They increase the growth rate of tra¢ c along a corridor and shift tra¢ c spatially. The spatial shift is small when the new highway is right next to the old route –tra¢ c clusters more around exits –but large when it is located miles away. From the perspective of local businesses that serve highway travelers such as gas stations (and potential entrants), the opening of a highway is an observable, anticipated change in the level of demand and sometimes demanders’locational tastes. Casual empiricism indicates obvious changes in industry structure that are associated with such changes: many highway exits have nearby service stations whose location can clearly be explained by the highway’s presence. The industry dynamics, however, are less clear: how large are the supply-side changes, along what margins do these changes take place, and what is the timing of these changes? These questions are the topic of this paper. We examine how industry structure adjusts to anticipated permanent demand shocks, and how the adjustment di¤ers depending on the extent to which locational tastes shift, by examining how the number and size distribution of service stations changes in hundreds of counties during the time surrounding the completion of Interstate Highway segments in these counties. We …nd that the timing and margin of adjustment of industry structure di¤ers, depending on whether the new highway is located close to or far from the old route. When the new highway is close to the old one, there is no evidence that the number of stations changes around the time it opens, but average station size increases by 6%, all of which takes place in the two years leading up to when the highway is completed. In contrast, when the new

2 highway is far from the old one (say, 5-10 miles), the number of stations increases by 8% but there is no signi…cant increase in average station size. Unlike the station size adjustment when the new highway is close, all of this increase takes place after the highway is completed. These results provide evidence on how this industry, which is characterized by high location-speci…c sunk costs, adjusts to demand changes. De- mand increases that have a limited spatial e¤ect are met by increases in station size, not additional stations. Our results indicate that this expansion –measured as increases in employees/station –takes place ahead of the demand increase. They are consistent with theories in which …rms have strategic investment incentives to preempt competitors, to the extent that increases in employees/station are correlated with irreversible investments (perhaps in new pumping capacity). In contrast, demand increases that are accompanied by spatial shifts are met primarily by new stations, not larger stations, and this happens only once demand increases. The di¤erence in the margin of adjustment is consistent with that pre- dicted by a broad class of industry structure models incorporating product di¤erentiation, such as Dixit and Stiglitz (1977) and Salop (1979), in which demand increases are met disproportionately by increases in the number of …rms in markets where entry would leave price-cost margins unchanged (such as when buyers are sensitive to spatial di¤erences) and by increases in …rm size when entry would lead price-cost margins to drop substantially. The di¤erence in the timing of the adjustment provides evidence against the proposition that spatial shifts –by opening new submarkets –increase …rms’ propensity to engage in pre-emptive entry (Spence (1977, 1979), Fudenberg and Tirole (1984), Bulow, Geanakoplos, and Klemperer (1985)) in this industry. This is interesting in light of the fact that opening a new service station near a new highway interchange involves industry- and location-speci…c sunk investments; such investments can function as credible commitments for …rms not to exit in the face of competition. Such a proposition ignores the possibility that uncertainty about demand or the competitive environment might be greater when demand increases lead new segments to open than when they do not, and thus the (real options-related) cost of pre-emptive capacity investments might be higher in such cases. (Dixit and Pindyck (1994)) Our results indicate that industry adjustment occurs later when demand shocks are accompanied by spatial shifts than when they are not, which is consistent with the view that spatial demand shifts increase the cost of pre-emption

3 relative to its bene…ts in this industry.1 Our analysis rests on our collection and combining of several data sets. These include (a) highly detailed data on when narrowly-de…ned Interstate highway segments opened, (b) county-level data from 1964-1992 describing the number, employment, and size distribution of service stations, and (c) hand-collected measurements of the distance between Interstate highways and the intercity routes they replaced. These data allow a far broader analysis of the e¤ects of Interstate highway openings on local industry structure than in the previous literature on this topic, most of which examines the long-run e¤ects of a small number of highway bypasses.2 Our analysis goes beyond these studies by examining evidence from a far larger sample, examining the margins and timing of adjustments, and by comparing situations where there was a large and small spatial e¤ect. These aspects a¤ord us not only the ability to estimate e¤ects more precisely, but also to shed light on how the adjustment process di¤ers with the degree to which demand increases are combined with changes in tastes. Our work is related to several lines of empirical work in addition to the "highway bypass" literature. Several recent papers, including Chandra and Thompson (2000), Baum-Snow (2007), and Michaels (2008) independently use the same highway openings data to investigate other issues such as the e¤ect of public infrastructural investments on output, the e¤ect of highways on suburbanization, and whether decreases in transportation costs lead to greater specialization. Campbell and Lapham (2004) use a similar empirical framework to ours to research how ‡uctuations in U.S.-Canada exchange rates –and thus temporary demand shifts –a¤ect the average size and number of establishments in various retail segments in U.S. counties bordering Canada. Finally, our use of rural areas to investigate industry structure is similar in spirit to Bresnahan and Reiss (1990, 1991) and Mazzeo (2002) (see also Campbell and Hopenhayn (2005) for an extension of this analysis to larger markets). We are able to examine industry dynamics in a way these papers cannot, because we are able to observe the number and size distribution of …rms over long periods, and how these change in response to demand shocks that are similar in nature but take place at di¤erent times in

1 It appears unlikely that zoning or other local political constraints explain this result: contemporary accounts indicate that there was little local government planning associated with real estate development at most highway interchanges, particularly those that were not located near existing downtown areas. 2 See for example, Texas Transportation Institute (1966).

4 di¤erent areas. Our data, however, limits our analysis. Like Bresnahan and Reiss (but unlike Berry (1992)), we have data on the number of producers but not their identities. This prevents us from investigating the details of the process through which …rms expand and contract their output. Our results indicate that further investigation of this process with …rm-level data is warranted. The rest of the paper is organized as follows. Section 2 presents the ana- lytical background to the paper. We summarize what monopolistic competition models conclude about how industries should adjust to demand shocks, and how the adjustment should di¤er according to the extent to which price- cost margins are expected to fall post-entry. We also discuss the bene…ts and drawbacks of capacity expansion, drawing from the pre-emption and real options literatures. Second 3 presents the institutional background, summarizing important trends in the industry between the 1960s and 1990s. This serves as the backdrop for our empirical analysis. Second 4 describes the data and shows aggregate relationships between the timing of highway com- pletions and changes in average service station size. Second 5 presents and discusses our main results. Section 6 concludes.

2 Industry Adjustment to Demand Shocks

2.1 Industry Structure Models A large class of models in industrial organization sheds light on how industry structure should adjust in the long run to permanent demand shocks.3 A general principle from this class of models is that increases in market size can lead either to more …rms or larger …rms, depending on the extent to which price-cost margins decrease as the number of …rms increases. In situations where price-cost margins do not change with entry, increases in market size will lead to more …rms, but not larger …rms.4 In contrast, if price-cost margins decrease with entry, increases in market size should tend to lead to larger …rms –industry adjustment will take more of the form of larger …rms than if price-cost margins do not change with entry. To illustrate this point, consider an industry with S identical potential demanders, each with demand q(p) for the industry’sgood, so that industry-

3 Spence (1976), Dixit and Stiglitz (1977), Salop (1979), Sutton (1991). 4 We are ignoring here the possibility of endogenous sunk costs a la Sutton (1991).

5 level demand is Q = Sq(p). Assume that there are a large number of potential suppliers, each of whom can produce at …xed cost F and marginal cost c. If N …rms enter, each faces a residual demand curve X = Sx(p; N), where x(p; N) is the number of units they sell to each of the S demanders. We assume that xp < 0, xN 0, xpN 0: …rms’residual demand curves are downward-sloping, residual demand (weakly) decreases with the number of competitors, and demand is (weakly) more price-sensitive the greater the number of competitors. These assumptions summarize demanders’tastes for …rms’ goods in this market, and therefore substitution patterns. For example, the second and third of these assumptions imply goods are (weak) substitutes. The cross-derivative xpN , which indicates the degree to which the slope of a …rm’s residual demand curve (per customer) changes with N is important to the analysis because it corresponds closely to how much equilibrium price-cost margins fall with N. A symmetric equilibrium in this industry satis…es:

p(x) + p0(x)x = c

p = c + F=Sx(p; N) Q = NX These equations imply that marginal revenue equals marginal cost for each …rm, price equals average costs for each …rm, and supply equals demand in the aggregate. The equilibrium is a triplet (p,x,N ) that solves these equations, subject to the expressions for industry- and …rm-level demand above. We are interested in how this equilibrium changes when S increases. An increase in S has no direct e¤ect on the …rst equation: …rms continue to produce at a point where their sales per customer x equates their marginal revenue per customer and marginal cost. Increases in S rotate …rms’residual demand curves outward, leading them optimally to sell more at the same price. However, an increase in S leads the right side of the second equation to increase relative to the left: …rms’average costs fall below price. At issue is how p, x, and N adjust to restore this relationship. First consider the case where xpN = 0: increases in N have no e¤ect on the slope of …rms’residual demand curves. This would be the case if the increase in market size elicited the entry of new products that are not substitutes to existing products. Then increases in N would a¤ect the second

6 equation only through x, and the condition p = AC would be restored at a point where X = Sx(p; N) was exactly the same as before. The number of …rms would increase, but quantity per …rm would not change. Next consider the case where xpN < 0: increases in N lead the slope of …rms’residual demand curves to be shallower, as would be the case if the increase in market size elicited the entry of new products that are substitutes. Now increases in N lead to decreases in equilibrium prices; in terms of the second equation, they a¤ect both the left side through p and the right side through x(p; N). The decrease in price, and therefore price-cost margins, implies that X will be greater in the new equilibrium than the old, because if price-cost margins fall, …rms must sell more units in order to satisfy the break-even condition p = AC. When increases in market size (potentially) elicit the entry of new products that are close substitutes, industry structure will adjust on di¤erent margins than in the case where such increases do not elicit the entry of close substitutes: adjustment will involve increases in …rm size, not just in the number of …rms. We apply this to our context straightforwardly. Other studies have shown that the opening of new Interstate highways increased travel along the corridor the highway serves.5 Suppose that this also increased the demand faced by service stations.6 Consider …rst situations where new highways are located on top of the previous route. New highway openings would primarily a¤ect tra¢ c patterns by forcing vehicles to get on and o¤ the road at exits, thus leading locations along the highway but between exits to have less tra¢ c. However, the e¤ect of this spatial change would be limited because service stations already tended to be located at important intersections between the previous road and other important roads, and the exits of the new highways were generally at these intersections. Such new highways would not lead to the creation of any new spatial segments: new entrants would be just as close substitutes to existing …rms as in the previous equilibrium. In contrast, new highways might lead to the creation of new spatial segments

5 Summarizing research on the impact of Interstate Highways on tra¢ c in corridors, Federal Highway Administration (1970) reports that "tra¢ c increases were steady...before opening, 3 to 5 percent annually. After opening, tra¢ c increases on the Interstate ac- celerated to annual rates of 10 percent and more for as much as 10 years after opening." The growth rate of tra¢ c through a county thus tended to increase after Interstates were completed. 6 Evidence below will suggest that it did: highway openings are associated with long-run increases in service station employment in our sample counties.

7 when they are located far from their previous route. Here they would have a more signi…cant e¤ect on tra¢ c patterns, leading service station demand to shift away from the old route and toward the new highway exits. Given this new set of spatial tastes, a new entrant could be a more distant substitute than it could have been given the previous set of spatial tastes. Its impact on price-cost margins would be less. It follows that the margin upon which industry structure adjusts should be more along the lines of the number of stations when highway openings shift demand spatially and more along the lines of the size of stations when they do not.

2.2 Pre-emption and Real Options Other literatures in which models are explicitly dynamic investigate …rms’ incentives with respect to the timing of new capacity additions or entry. A broad lesson of these literatures is that there can be strategic bene…ts from expanding or entering before competitors do, possibly ahead of demand shocks, but there are option-related costs of doing so. The pre-emption literature focuses on the bene…t side of the ledger: …rms’ strategic incentives for capacity expansion.7 The main idea is that …rms can bene…t from expanding capacity or entering ahead of competitors to the extent that doing so weakens competition ex post. The logic, as applied to capacity expansions, is similar to that in Stackelberg games. If a …rm is able to commit to expanding capacity or entering a market, and this diminishes (potential) competitors’ marginal returns to investment or entry, this will lead to less investment or entry by competitors. This, in turn, bene…ts the preempting …rm by leading it to face softer competition on the equilibrium path.8 Note that for this logic to go through, it is necessary that (a) capacity expansions or entry involve irreversible market- and industry-speci…c investments that commit capacity to stay in the market irrespective of what competitors do, and (b) capacity expansions or entry diminish competitors’ marginal returns to capacity additions or entry, perhaps by ensuring that price-cost margins would be low if competitors expanded or entered.

7 This literature includes, for example, Spence (1977, 1979), Fudenberg and Tirole (1984, 1986), and Bulow, Geanakoplos, and Klemperer (1985). 8 Pre-empting …rms could also obtain competitive advantages, for example from better locations, that would persist in the face of entry. If so, the analysis is similar but provides for additional strategic incentives.

8 It follows from this logic that if …rms foresee a positive demand shock and have the opportunity to expand ahead of demand, they may have an incentive to do so in order to preempt their competitors. Although this has short-run costs –their pro…ts before the demand shock are lower than they otherwise would be –it may have long-run bene…ts to the extent that it provides them a competitive advantage or weakens price competition in the future. It also follows that pre-emption incentives might be larger when demand shocks are associated with the opening of new spatial segments than when they do not, because pre-emption should have a greater marginal e¤ect on ex post price-cost margins in such circumstances. This is analyzed explicitly in Fudenberg and Tirole’s(1986) model of spatial preemption. These authors’ analysis illustrates why both incumbents’and new entrants’preemption incentives are greater if they are able to enter parts of product space away from incumbents’existing capacity than if they are not. The reason is simple: an objective of preemption is to soften ex post competition, and the marginal e¤ect of capacity additions (or entry) on other …rms’capacity decisions (or entry decisions) will be greater in areas where …rms have not yet made capacity commitments than in areas where they have done so. The real options literature (Dixit and Pindyck (1994)) highlights the cost of capacity commitments. The general idea is that while such commitments can have the strategic bene…ts described above, they have option-related costs: making commitments now forecloses the option of making commitments later instead. These costs are greater in situations where there is greater economic uncertainty, because maintaining options is more valuable in such circumstances. This can lead …rms optimally to delay capacity expansions or entry, relative to situations where there is less uncertainty. These two literatures inform our empirical analysis in the following way. One implication is that if capacity investments or entry do not involve industry- and market-speci…c sunk costs, then there are neither strategic bene…ts nor option-related costs that a¤ect how …rms adjust to demand shocks. The timing of adjustment should be independent of these forces. Absent other factors the adjustment should coincide with the demand shock, irrespective of whether there is a spatial demand shift.9 Thus, …nding that the adjustment

9 The quali…er here is important: other factors might be important in our context. For example, zoning or other regulatory factors may make it more di¢ cult to expand or enter in some circumstances than others. We will investigate this possiblity in more detail in future drafts. Our current under-

9 coincides with the demand shock irrespective of whether there is a spatial demand shift is consistent with the proposition that investments and entry do not involve industry-speci…c sunk costs. A second implication is that it is a priori unclear how the timing of adjustment to demand shocks should di¤er, depending on whether the demand shock is accompanied by a spatial shift in demand. On one hand, the strategic pre-emption related bene…ts might be greater, and this would lead the adjustment to begin sooner than when there is no spatial shift. On the other hand, demand uncertainty may be greater when there is a spatial shift –for example, …rms might be uncertain regarding the extent to which demand from local "non-through" tra¢ c will shift once the highway is completed. If so, this would o¤set the pre-emption e¤ect, and may lead industry adjustment to take place later than when there is no spatial shift. Our empirical work will provide evidence on whether the timing of adjustment di¤ers with the extent to which highway openings involve a spatial shift. Finding that the adjustment takes place sooner when there is a spatial shift provides evidence supportive of the hypothesis that the preemption-related strategic incentives are strong relative to the uncertainty-related strategic costs. Finding instead that the adjustment takes place later in such circumstances provides evidence supportive of the hypothesis that any strategic- related bene…ts are more than o¤set by the option-related costs.

3 Service Stations

We …rst report general trends with respect to service stations during and slightly outside our 1964-1992 sample period. The numbers are as reported by the U.S. Census in either County Business Patterns or the Economic Census (as part of the Census of Retail Trade or, before 1972, the Census of Business).

3.1 General Trends Figure 1 presents several series that track the number of service stations in the U.S., and subsets thereof. The top set of points represents all service standing is that zoning and regulation played much less of a role in these markets in the 1960s and 1970s than they currently do (in part because environmental regulation was less restrictive), but we do not fully understand their e¤ect at the time.

10 stations. It shows that the number of service stations increased slowly during the 1960s and early 1970s, growing by 7% from 1963 to its 1972 peak of about 226,000. This number decreased sharply during the 1970s and early 1980s, falling by more than one-third to about 135,000 in 1982, and has been relatively stable since then. The …rst third of our sample period is one in which new station openings were exceeding closings, but service stations were, on net, exiting the market during most of our sample period. The second series tracks the number of service stations with positive payroll; the di¤erence between this and the …rst series represents stations with no employees: these are stations where the owner or owners operate the station by themselves. This di¤erence shows that nonemployer stations became increasingly rare starting in 1972, falling from 43,074 in 1972 to 19,326 in 1982. But about three-quarters of the overall decline in the number of stations is accounted for by the 67,000-station decrease in the number of stations with employees. The fact that the general trends above appear as well when looking only at such stations is worth noting because our main data source tracks only stations with employees. The other series track the number of "reporting units," as published in County Business Patterns (CBP). The county-level data that we analyze later is from this source. There is a break in this series because the de…nition of a "reporting unit" changed in the middle of our sample period.10 Starting in 1974, the CBP "reporting unit" is the establishment –in this context, the service station –and the numbers published in the CBP track those published in the Economic Censuses (EC) closely. But before 1974, the de…nition of a "reporting unit" was such that …rms operating multiple service stations in the same county reported these stations as a single observation; the county-level data therefore reported the number of …rms competing in the county, not the number of service stations. Time series of CBP data before 1974 re‡ect not only the entry and exit of single-station …rms, but also any combinations or spin-o¤s of service stations within the same county. Comparing the reporting unit counts and the establishment counts before 1974 indicates the degree to which …rms operated multiple stations in the same county: the number of establishments with payroll exceeds the number of reporting units by 10-12% in 1963 and 1967, but by 25% in 1972. This provides evidence that, starting

10 This change corresponded to a change in how the Internal Revenue Service asked …rms to report employment and payroll data. There was also a change in the employment size categories the Census used. Before 1974, the three smallest categories were 1-3, 4-7, and 8-19 employees; after 1974, these were 1-4, 5-9, and 10-19 employees.

11 in the late 1960s, it became increasingly common for …rms to own multiple stations in the same county. The size and composition of service stations changed during our sample period. Figure 2 reports time series on average employment size. The EC series show that the average employment size of service stations grew throughout our sample period, increasing by about 125% between 1964 and 1992. Turning to the CBP-derived series, the employment size of the average reporting unit –that is, average within-county …rm size –increased by 41% between 1964 and 1972. Employment per station with payroll increased by about 30% during this time; hence, only about one-fourth of the increase in within-county …rm size re‡ects increases in the number of stations per …rm within counties rather than increases in the number of employees per station. Although it will be important in our main analysis to account for and investigate the degree to which the pre-1974 data re‡ect …rm-level rather than station-level phenomena, overall the bulk of pre-1974 employment size increases appears to re‡ect increases in station size. Other Census …gures published on a consistent basis since 1972 show corresponding increases in size; we depict these in the …rst few columns of Table 1. Gallons per station increased steadily between 1972 and 1992, more than doubling during this time. This re‡ects both an increase in the number of gallons per pump, which grew by 63%, and the number of pumps per station, which grew by 37%. The increase in pumps per station during this period occurred entirely between 1977 and 1987. Employees per pump grew only slightly, and was almost constant between 1977 and 1992. These …gures indicate that at the same time average employment per station was increasing, stations’pumping capacity was increasing, and this pumping capacity was being utilized more intensively. The rest of Table 1 depicts two well-known changes in service stations that occurred duirng this time. One is the movement toward self-service. This began in the early 1970s, and the share of sales that are self-service exceeded 90% by 1992. The other is the change in service stations’ancillary services away from automotive services and toward convenience stores. These changes did not entirely coincide. The movement away from automotive services began in the early 1970s and was essentially complete by 1982; the share of service station revenues from tires, batteries, and accessories declined from 10% to 3% during this time, and has remained low ever since. In contrast, the increase in the revenue share of convenience store items –food, alcohol, and tobacco –occurred predominantly after 1982; the

12 revenue share from these categories increased from 5% to 15% between 1982 and 1992, and has increased since then to about 25%.11 This study focuses on periods surrounding when Interstate highways were being completed, and the phenomena we uncover mainly re‡ect changes in the number and size distribution of service stations that occured during the 1960s and 1970s. The di¤usion of self-service gasoline and the diminishing importance of auto repair occurred during this period, but the rise of convenience store-service stations took place later. Importantly, increases in the employment size of service stations pre-dates the rise of such stations, and coincides at least in part with the decline of the provision of auto-related services and the increase in self-service –two trends that would tend to decrease the use of labor. The increase in service stations’employment size therefore likely re‡ects some combination of (a) stations being physically larger, as manifested in more pumps, and (b) stations being open longer hours. The former is likely to be particularly important starting after 1977, when pumps/station but not gallons/pump was increasing; the latter is likely to be particularly important between 1972 and 1977 when the reverse was true.

4 Data

4.1 Description Our two primary sources of data provide information about highway openings and local market structure for service stations. Our data on highway openings come from the U.S. Department of Trans- portation’s "PR-511" …le. These data describe the milepost, length, number of lanes, pavement type, and opening date of segments of the Interstate Highway System that were open by June 30, 1993 and built using Interstate Highway funds. The data cover nearly the entire System.12 Highway segments in these data range in length, but the vast majority are less than …ve

11 A third change during this period was the movement from leaded to unleaded gasoline. This, like self-service, began in the early 1970s and was essentially complete by 1992. Many stations o¤ered both leaded and unleaded gas by o¤ering them at di¤erent pumps or islands; existing stations often replaced a pump that supplied leaded premium with one that supplied unleaded regular. 12 A small fraction of the IHS includes highways that were not built with Interstate High- way funds, but were incorporated into the System later. (I-39 in Illinois is an example.) These highways are not in our data.

13 miles long and many are less than one mile long. Opening date is described as the month-year in which the segment was open for tra¢ c. The milepost and length variables in the PR-511 indicate where the highway segment is located along the route. We hand-merged these variables with geographic mapping data from the National Highway Planning Network to identify the county in which each of the PR-511 segments is located.13 This produced a highly-detailed dataset on the timing and location of Interstate Highway openings. We then aggregated these data up to the route-county level. For each route-county (e.g., I-75 through Collier County, FL), we calculated the total mileage within the county, the total mileage completed by the end of each calendar year, and the share of mileage completed by the end of each calendar year. Highways were normally completed in stages, so it is not unusual for a route to be partially complete within a county for some period of time, then fully completed within the county a few years later. This cumulative share variable, csmiit, is a key independent variable in our analysis. We also develop a corridor-level version of this variable, ccsmiit, which accounts for the possibility that tra¢ c volumes in a county are not only a¤ected by highway openings in the county, but are also a¤ected by highway openings in other counties along the same tra¢ c corridor. For example, tra¢ c in Boone County, Missouri is not only a¤ected when Interstate 70 was completed in Boone County, but also when it was completed in other counties between Kansas City and St. Louis. We describe the details of how we de…ne corridors and how we assign highway segments to corridors in the Appendix. The basic idea is simple, however. Most corridors are de…ned as highways that connect two central cities with at least 100,000 population; Interstate 70 between Kansas City and St. Louis is an example. For each corridor, we calculate the share of Interstate Highway mileage completed in each year, and assign this variable to each county that lies along the corridor; for example, we calculate the share of Interstate 70 between Kansas City and St. Louis that was opened in each year, and assign this variable to each county through which I-70 passes between these two cities. We utilize tra¢ c count data on Interstates from the U.S. Department of Transportation’s Highway Performance Monitoring System (HPMS), to

13 These data are maintained at: http://www.fhwa.dot.gov/planning/nhpn/. The PR- 511 …le contains a variable that indicates the county in which the segment is located, but other researchers (Chandra and Thompson, 2000) have noted that this variable contains errors. We use the PR-511 data in checking our construction of this variable.

14 construct a variable that measures the amount of through tra¢ c within each of our corridors. These allow us to develop a variable that distinguishes among counties by whether demand from through tra¢ c is important relative to demand from locals. We develop a measure of through tra¢ c in the county by taking the minimum daily tra¢ c count on the Interstate within each of our corridors and assigning it to each county in the corridor. Call 14 this thrui. We then construct a variable thrusharei = thrui=(thrui+ empi) where empi is the county’s 1992 employment. The mean value of thrusharei across our 677 counties is 0.55. The maximum value is 0.97, which is in Culberson County, TX –a very small county on a fairly heavily traveled stretch of Interstate 10 in west Texas. The minimum value is 0.01 in Kennebec County, ME, the largest county on the corridor which includes the least-traveled stretch on the Interstate Highway System (the northernmost part of I-95). We augment these data with a measure of how far the Interstate highway shifted tra¢ c. We did this via the following procedure. Using mid-1950s road maps, we …rst designated the route each segment of Interstate highway likely replaced (the "old route"). The general procedure was to look …rst at the major cities that the current Interstate connects, then assess the most direct major route between these cities as of the mid-1950s. For example, the "old route" for I-95 between Boston and New York is US1. Often, establishing the old route is more di¢ cult because the old route either no longer exists or is a minor road. The "old route" for I-5 in Oregon is old US99, which in many places currently exists as a minor road adjacent to I-5. Once the "old route" was established, we measured the "crow ‡ies" distance between each current Interstate exit and the old route. This was done using Google Maps and ancillary tools. Finally, we averaged this distance across the exits within each route-county. This produces a variable disti (or "distance from old route") that characterizes the spatial shift in tra¢ c brought about by the Interstate highway. This measure ranges from zero for many route-counties (where the Interstate merely was an upgrading of the old route) to over 20 miles. The median value in our sample is 1.25 miles;

14 The HPMS data provide tra¢ c counts measured periodically on the Interstates in our sample. These data are reported consistently starting in 1993. Our measure uses data from 1993-98 to reduce noise in the counts due to sample sizes. We constructed thrui by …rst calculating the minimum tra¢ c count on the route*county in each of these years, then averaging this quantity across these years. We then took the minimum value of this county-level average across the corridor.

15 the 25th and 75th percentile values are 0.5 and 3.0 miles, respectively. Our data on local market structure for service stations come from County Business Patterns, published annually by the U.S. Bureau of the Census since 1964. CBP contains county-level data on narrowly-de…ned industries, including "gasoline service stations," SIC 554. We obtained these data in electronic form from 1974-1992; we hand-entered these from published reports from 1964-1973. For each year and county, these data report employment and payroll in the industry within the county. They also report the total number of service stations and the number in several employment size categories. We describe above how the reporting unit and size categories reported in CBP changed starting in 1974. Our data contain missing values for some county-years, especially in the very smallest counties. Missing values arise for industry employment and payroll when the Census deems that publishing these would disclose con…- dential information regarding individual …rms. Such disclosure issues do not arise for the local industry structure variables; these are considered publicly- available information in any case. However, to economize on printing costs, the Census did not publish these data for industry-counties with small numbers of employees (typically fewer than 100); they are available only in electronic versions of the data. We therefore have missing values for these variables in very small counties, particularly in years before 1974. The CBP data form our dependent variables, the most important of which are the number and employment size distribution of gas stations (before 1974, …rms) within the county in an particular year. The bulk of our analysis relates these variables to the timing of highway openings.

4.2 Sample Criteria Our analytic framework anticipates using highway openings to represent spatial shifts in demand for gasoline, and envisions contexts where these shifts are uncomplicated: for example, a situation where a new highway opens that parallels an existing road that had previously served both local tra¢ c and "through" tra¢ c. This is unreasonable in urban contexts, since one would expect the spatial distribution of demand for gasoline to be less dependent on the location of the most important "through" roads. We therefore con- duct our analysis on a part of our sample that includes only less dense areas where tra¢ c patterns are relatively uncomplicated. First, we use only counties with a single two-digit Interstate and no three-digit Interstates; this is

16 a simple way of eliminating most large cities as well as other counties with complicated tra¢ c patterns. We then eliminate all counties where 1992 employment exceeds 200,000 because some populous counties remain after this cut (for example, New York, NY). We also eliminate all counties through which the highway passes but there is no exit; most of these are cases where the highway clips the corner of a county. Finally, we employ our main analysis on a "balanced panel" which includes only counties where the number of service stations is nonmissing in each year between 1964-1992. Our main sample ultimately includes 677 counties; we depict these counties in Figure 3. This map indicates that our sample counties come from all over the United States, tracing the non-urban parts of the Interstate High- way System. Di¤erences in the shading of these counties indicate di¤erences in when the highways were completed; broadly, they were completed later in west than in other regions of the country. In addition, di¤erences in the shading of the highway indicate counties where the new highway was far from the previous intercity route, de…ned here as farther than 3 miles. It was more common for western Interstates to be completed close to the previous route than Interstates in other areas of the country, in large part because the population is less dense in the west than in east or south.

4.3 Patterns in the Data Table 2 presents the timing of "two-digit" Interstate Highway completion as reported in the PR-511 data, and for our balanced panel counties. From the left part of the table, 20% of two-digit highway mileage was open by the end of 1960; most of this mileage consisted of toll roads in the east that predated the Interstate Highway System (such as the Pennsylvania Turnpike) and were incorporated into the System once it was established. About 55% of two-digit mileage in the System was completed during the 1960s; the peak construction year was 1965. Another 20% was completed during the 1970s, and the …nal 5% thereafter. The counties in our balanced panel account for 18,833 miles of Interstate Highways, about half of the two-digit mileage in the System as a whole. The timing of highway construction in this subsample sample mirrors that of the system as a whole, peaking in the mid-1960s, then steadily declining during the years that followed. The timing of Interstate Highway construction means that our analysis will center on events that mostly took place in the 1960s and early 1970s, and our creation of a dataset that examines changes in industry structure during this time exploits this.

17 Table 3 presents time trends in the number and size distribution of …rms (starting in 1974, service stations) in our 677 balanced panel counties. The trends in these counties are very similar to those in the U.S. as a whole. The number of …rms/county fell slightly between 1964 and 1973, and the number of service stations per county fell by about one-third between 1974 and 1992. The right four columns of this table indicate changes in industry structure and depict the movement toward fewer, larger …rms and service stations; there are steady decreases in the number of businesses in the smallest size category and increases in the number of businesses in the other size categories. Table 4 presents some initial evidence on whether the timing of industry structure changes are related to the timing of highway openings. We place counties into three categories according to the year the highway was completed in the county: 1965 or earlier, 1966-1971, and 1972 or later. We then calculate employees per …rm (starting in 1974, per station) within these categories.15 Table 4 indicates that average …rm size was similar across these categories in 1964; in each, there were about three employees per …rm. Em- ployment size increases steadily during this period; in 1992, the average gas station in each of these county classes had roughly seven employees. But the timing of this increase di¤ered across these categories. Firm size increased in the "early" counties relative to the "late" counties early in our sample; by the early 1970s, the di¤erence was about 10%. The opposite was true late in our sample, after the mid-1970s, average station size increased in the late counties relative to the early counties. Figure 4, which depicts the ratios between the "late" and "early" counties each year, shows this pattern. This evidence indicates that increases in the size of service stations corresponded to the completion of Interstate highways. Figure 5 contains further detail with respect to changes in industry structure. Here we report late/early ratios for the number of …rms (or stations) per county for di¤erent employment size categories. Looking at these provides further information regarding the dynamics of changes in industry structure; we know from Table 4 that the general pattern is one of a decrease in small businesses and an increase in larger ones; here we investigate whether both of these changes correspond to highway completion. Figure 5 indicates that there is a di¤erence in the time pattern between the small and the other size businesses. The late/early ratio decreases for the "large" cate-

15 The quantites in Table 4 and Figure 4 use only counties where we observe service station employment in each year, N=470.

18 gory through 1974, then increases steadily throughout the rest of the sample. This indicates that the number of large …rms increased more in the "early" counties than the "late" counties early in our sample, while the opposite was true at the station level later in the sample. A similar pattern appears for the mid-sized businesses. The increase in the number of large and mid-sized businesses took place earlier in counties where the highway was completed earlier. We do not observe such a pattern, however, when looking at the smallest size category: the number of small businesses declined at about the same rate in counties where the highway was completed early as late. This evidence suggests something interesting about industry dynamics during this time. The Figure 4 relationship between the timing of highway construction and increases in employment size does not appear to be driven by a mechanism in which new highways lead to increases in the number of large …rms or stations and corresponding decreases in the number of small ones. Small stations are exiting the market throughout our sample period, but there is no evidence that changes in the number of small stations are related to the timing of highway construction. Instead, this relationship is consistent with models in which new highways lead to increases in the number of large stations without changing the number of small ones. Below we will see this again in the econometric analysis and interpret the result in light of sunk costs and exit patterns.

5 Empirical Model and Results

Our empirical speci…cations follow Campbell and Lapham (2004). We estimate vector autoregressive speci…cations of the form:

yit = i + t + yit 1 + xit + "it

In the …rst set of results that we will present, yit is a 2 1 vector containing the logarithms of the number of service stations per capita in county i at time t (nit) and their average employment (ait). We used the county’spop- ulation in 1980 for nit’sdenominator, and before 1974 nit equals the number of …rms per capita. The vector xit contains our highway opening variables, including up to three leads and lags; we describe this part of the speci…cation in more detail below. The parameters i and t are county-speci…c and year-speci…c e¤ects. The parameter i represents time-invariant factors that lead the number and size of service stations to di¤er across counties,

19 and t embodies trends and aggregate ‡uctuations that a¤ect all counties’ industries equally. Removing these county-speci…c and time-speci…c e¤ects isolates the changes in the number and sizes of service stations around the time of Interstate highway openings relative to the county’sown history and national developments. The speci…cation’s autoregressive structure allows the impact of an Interstate’sopening to occur gradually. The coe¢ cients of 1 give the initial impact, while (I ) measures the long-run change. Setting aside for now leads and lags, the vector xit includes up to three highway opening variables: ccsmiit, csmiit, and csmiit disti. Including ccsmiit accounts for the possibility that the level of demand for gasoline in a county depends on corridor-level construction; the interaction csmiit disti allows for the possibility that the e¤ect of the completion of a highway in given county has a di¤erent impact on local industry structure, depending on the size of the spatial shift in demand. One would expect the impact of new highway openings on local industry structure to di¤er, depending on whether through tra¢ c accounted for a small or large share of gasoline demand relative to the demand from locals. In some speci…cations we therefore interact (1 thrusharei) and thrusharei with xit. We think of the coe¢ cients on the (1 thrusharei) interactions as re‡ecting the extent to which highway openings a¤ect local industry structure by changing the tra¢ c patterns of locals, and the coe¢ cients on the interactions wtih thrusharei as re‡ecting the extent to which they do so by changing through tra¢ c. Although it is reasonable to think that the former e¤ect could matter –for example, highway construction could increase locals’ gasoline demand by decreasing their travel costs –the latter e¤ect captures much of the spirit of the paper. Much of our focus will therefore be on the coe¢ cients on the interactions with thrusharei. This draft reports results when we estimate our speci…cations using OLS. Future versions will use estimators that account for the econometric endo- geneity of yt 1. Based on our experience with these estimators as applied to CBP data, we expect the results to change little when we do so.

20 5.1 Basic Results 5.1.1 Number and Average Size of Stations

16 Table 5 presents results from several speci…cations. In the top panel, xt contains no leads or lags, and includes only csmiit. Looking …rst at the autoregressive coe¢ cients, all are positive and signi…cant: the impact of shocks to the number and average size of service stations in a county is therefore distributed over time. The highway opening coe¢ cient is economically and statistically zero for the number of stations, and is positive and signi…cant for the average employment size of stations.17 The magnitudes of the highway opening coe¢ cients, combined with the autoregressive coe¢ cients, imply that the opening of a highway is associated with no change in the number of …rms, but a 6% long run increase in the average employment size of service stations in the county, one-third of which (1.9%) occurs in the year that the highway opens. The second panel adds a lead and lag to the highway opening vector. The main result is the positive and signi…cant coe¢ cient on the "-1 year" coe¢ cient in the average employment size regression: the increase in average size of service stations begins before the highway opens. The sum of the lagged coe¢ cients is approximately unchanged. The …nal two panels extend the analysis to two and three leads and lags. While the autoregressive coef- …cients and the sum of the lagged coe¢ cients –and thus our estimate of the long-run impact of highway openings –are approximately the same as in the other panels, the individual highway openings coe¢ cients are estimated with more noise. The positive estimates of the "zero, one, and two years before" coe¢ cients suggest that average station size increased before opening; the coe¢ cient on the "one year after" coe¢ cient indicates that it fell somewhat the year after the opening.

16 All speci…cations allow the autoregressive coe¢ cients to vary for the year 1974, to account for the change in the Census de…nition of reporting units. We have also estimated speci…cations that allow these coe¢ cients to vary before and after this change, and to vary in each year. The estimates on our highway openings coe¢ cients vary little when we do so. 17 Before 1974, the unit of observation in the data is the "county-…rm." To avoid con- voluted language, we will use the term "station" to refer to our unit of observation before and after 1974. This will be supported by empirical evidence that we present below: the results do not appear to di¤er before and after 1974, suggesting that highway openings were associated with changes in the number and size of stations rather than stations’ propensity to be part of multiestablishment …rms.

21 Figure 6 presents impulse-response functions for highway openings on ln(ait) that are implied by these coe¢ cient estimates. The speci…cations with two and three lags indicate that average service station size increases by about 5% in the two years leading up to a highway opening, decreased by 2-4% in the year after opening (though this is not statistically signi…cant), and increased again thereafter. All speci…cations indicate a statistically signi…cant long-run increase of 5-6%. While we …nd these general results interesting, these speci…cations do not di¤erentiate between highway openings with small and large spatial demand shifts. Below we …nd that once we do, the industry dynamics become clearer.

5.1.2 Size Categories Table 6 presents more detail regarding these patterns by looking at how the number of stations in our size categories changed around the time of highway openings. This table reports results where the dependent variable yit is a vector of the number of stations in each of the four employment size categories reported in Table 3. For brevity, we show results only for zero to two leads and lags; the three leads and lags speci…cation produces results similar to the two leads and lags one. The main result in this Table is that the patterns in Table 5 and Figure 6, which depict increases in average station size, are accounted for by a signi…- cant increase in the average number of "large" stations with 8-19 employees (or, after 1973, stations with 10-19 employees). Figure 7 plots the impulse- response function for highway openings on the number of stations in this category. Our estimates indicate that the number of large stations increased by 0.8 stations during the two years leading up to the highway opening, and in the long run increased by 1.2-1.4 stations. This is fairly large relative to the sample mean of 3.2, and about 1/3 of the average increase in the number of such stations between 1974-1992. In contrast, there is neither evidence of net entry or exit in the other size categories.

5.1.3 Do These Patterns Di¤er After 1973? We next investigate whether our estimates of the relationship between highway openings and industry structure change after 1973. By doing this, we examine several hypotheses. One has to do with whether the patterns we uncover re‡ect …rm-level or station-level e¤ects. Recall that our data are

22 reported at the …rm level rather than the station level until 1974. One interpretation of the results in Table 5 is that owners of existing service stations in the county (perhaps the upstream re…ner) may have added another station around the time that the highway opened in the county. If so, our results would re‡ect changes in the size distribution of …rms but not stations. Finding that the e¤ects we uncover are signi…cantly weaker after 1973 would provide evidence that our results re‡ect the growth of chains not stations; in contrast, …nding no di¤erence in these e¤ect would provide no evidence in favor of this hypothesis. A second reason for such a test is that, as Table 1 indicates, service stations changed starting around this time –self-service stations became more prevalent, and later on, service stations started to have convenience stores. Finding that the results we uncover are stronger after 1973 would provide evidence consistent with the hypothesis that the changes we uncovered are interrelated with changes in stations’ format associated with self-service or convenience stores. Finding no di¤erences would provide no evidence consistent with this hypothesis. Results are in Table 7. In short, the "pre-1974" coe¢ cient estimates for look very much like the overall estimates, and there is no evidence of a signi…cant change in this vector after 1973. For each speci…cation and each equation, we fail to reject the null that the change in the vector is zero, using Wald tests of size 0.05. To some extent, this re‡ects the simple fact that over 3/4 of two-digit Interstate highway mileage (both overall and in our subsample) had opened by 1973. However, enough mileage was constructed after this time so that the test has some power, and …nding no signi…cant changes provides evidence that Interstate highways were having a similar impact on local service station market structure before and after this time. We …nd no evidence that our results re‡ect only the expansion of multiestablishment …rms, or are driven by changes in station format.

5.1.4 Discussion The estimates to this point indicate that on average, local markets adjusted to highway openings through increases in average station size, not in the number of stations, and that this adjustment began two years ahead of the highway’sopening. A manifestation of average station size is in the increase in the number of large stations, which may either re‡ect the entry of new stations or a signi…cant expansion at some existing stations. They provide a preliminary indication of the industry dynamics associ-

23 ated with Interstate highway openings. On average, the margin of adjustment is on the intensive margin rather than the extensive margin: station size increased, but there is no evidence that the number of stations did. In- creases in station size began well ahead of the year highways open, evidence suggestive of preemptive capacity additions. The estimates also suggest that sunk costs shape industry dynamics. Re- call that during our sample period, the number of large stations was increasing and the number of small stations was decreasing. Our results indicate that, at least during the time window that we investigate, highway openings are associated with an increase in large stations but there is no evidence that highway openings are associated with a decrease in the number of small stations. This fact is what one might expect in an industry where there are signi…cant industry-speci…c sunk costs –the fact that it is costly to convert a service station to other purposes would lead exit to be relatively insensitive to demand shocks and competitive conditions.18 These patterns, while interesting, mask important di¤erences in the margin and timing of adjustment between situations where the new highway was close to and far from the old route. We present and interpret evidence on these di¤erences in the next section.

5.2 Highway Openings, Spatial Demand Shifts, and In- dustry Dynamics We next extend the analysis by examining how the relationship between highway openings and industry structure di¤ers, depending on how far the Interstate is from the previous major route. We …rst run a series of simple speci…cations to examine whether the margin of adjustment di¤ers with how far the new Interstate is from the old route, and if so whether any e¤ects we …nd are nonlinear in distance from old route. Results are in Table 8; these are analogous to those in the top panel in Table 5 that include no leads or lags. We report here only the coe¢ cients on the cumulative share of miles completed in the county and interactions between this variable and "distance from old route." The estimates in the top panel indicate that highway openings are associated with

18 It is also what one might expect from watching the movie "Cars:" after all, the Radi- ator Springs service station had not yet exited the market, even though there apparently had been no through tra¢ c in the town for many years.

24 a greater increase in the number of stations when the Interstate is farther from the old route. The estimate on the interaction in the second column is positive and signi…cant. In the third column, we allow the distance e¤ect to be nonlinear by including an interaction with the square of distance; the estimate on this coe¢ cient is negative, but is not statistically signi…cant. Within the range of our data, the linear and quadratic speci…cations have similar implications: no evidence of a relationship between highway openings and changes in the number of …rms when "distance from old route" = 0, but a relationship that gradually increases in magnitude to about 0.025 as "distance from old route" increases to 10 miles (which is the 95th percentile "distance from old route" in our data). The bottom panel shows analogous results when examining the average employment size of service stations. In contrast to the top panel, there is no evidence of an e¤ect that di¤ers with "distance from old route." The long run increase of 5-6% we report above holds irrespective of distance. Combined, these speci…cations indicate a systematic di¤erence in how these local markets adjust to demand shocks: when the spatial demand shift is minimal, the industry adjusts through changes the average size but not in the number of stations. In contrast, when there is a signi…cant spatial shift, it adjusts through the number of stations as well. These patterns are consistent with the view of the broad class of models described in Section 2: demand increases without taste changes primarily lead to increases in average …rm size, but demand increases that are accompanied by taste shifts toward previously uncovered areas in "product space" are absorbed by increases in the number of …rms. Table 9 shows how the timing of adjustment varies with the magnitude of spatial demand shifts. These results are from speci…cations that include leads and lags, and allow the highway opening variables to interact with "distance from old route." In addition to the coe¢ cient estimates, we show estimates of the sum of the leads and lags, evaluated at distance = 0 and distance = 10, in the right part of the table. The main …nding from these speci…cations is that the timing as well as the margin of adjustment is di¤erent when comparing situations where the Interstate was near or far from the old route. This is suggested by the coe¢ cient estimates in the middle panel: in particular, by the positive and signi…cant coe¢ cient estimates on the "+1 year" interaction in the number of stations regression and on the "-1 year" coe¢ cient in the average station size regression. But it can be seen more easily in the impulse- response functions associated with these speci…cations, which we display in

25 Figure 8 and which use results from the middle speci…cation. In each of these, the three lines represent impulse-response functions evaluated at three distances: 0 miles, 1.25 miles, and 10 miles; these are at the 5th, 50th, and 95th percentiles of the distance distribution in our sample. The functions for 0 and 1.25 miles are similar: there is little change in the number of stations, but an increase in the average size of 6% during the two years leading up to the highway opening. Thereafter, the average size levels o¤. The function is much di¤erent for 10 miles. There is an increase in the number of stations of about 8%, starting after the highway is complete, but no signi…cant increase in the size of stations. These patterns provide evidence that the timing of the adjustment process di¤ers depending on whether there is a spatial demand shift: it happens earlier when the spatial demand shift is small than when it is large. When there is a limited spatial demand shift, average station size increases in the period leading up to year highway segments are completed, but there is no increase in the number of stations, and this increase occurs before the demand changes. When there is a spatial shift, the number of stations increases and this increase occurs after segments are completed. In Section 2 we highlighted a central trade-o¤ …rms face when responding to anticipated demand shocks in industries where capacity additions involve industry-speci…c sunk costs. Adding capacity earlier can have strategic pre-emption related bene…ts, but can have the drawback of foreclosing the option of not to invest. We discussed how one would expect both the bene…ts and drawbacks to be greater when demand shocks open new segments than when they do not: pre-emption might be more attractive in new segments where there are not existing competitors, but uncertainty might be greater to the extent that demand and competitive conditions are harder to forecast. Our results here indicate that industry adjustment to highway openings is systematically slower when openings create new spatial segments than when they do not. Thus, our evidence is consistent with the proposition that, while the pre-emption-related bene…ts from expansion may be greater when spatial demand shifts are greater, this is more than o¤set by the e¤ect of greater uncertainty.

26 5.3 Does The Timing of Adjustment Re‡ect Highway Openings In Other Counties in the Corridor? The results above indicate that the adjustments to industry structure are earlier when new highway openings involve a small spatial demand shift than when they involve a large one, and that these adjustments take place before the highway opens when there is only a small spatial demand shift. One interpretation of the latter result is that it re‡ects highway openings in other counties along the same corridor: the demand for gasoline in a county may increase before the highway in the county is completed because the highway has been completed elsewhere in the corridor, and this has led tra¢ c in the corridor to increase. If so, the latter result would not be evidence of pre- emption. We investigate this by including ccsmii in our speci…cation. Table 10 shows the results. The top panel shows speci…cations with no leads and lags. The coe¢ cient on ccsmii in the number of stations regression is economically and statistically zero: the results are essentially the same as in our base speci…cation. The story is somewhat di¤erent in the station size regression. The point estimate on csmii declines to 0.016 (down from 0.020 in the base speci…cation) and becomes not statistically signi…cant; the point estimate on ccsmii is 0.014 and not statistically signi…cant. This speci…cation indicates that it is di¢ cult to separately identify the impact on station size of highway openings in a county and in a corridor. The bottom panel, however, provides evidence that our pre-emption result does not re‡ect highway openings in other counties. Here we include a lead and a lag. We …nd that the results on csmii are almost identical to those in Table 9; in particular, there is a positive and signi…cant coe¢ cient on the one-year lead that is nearly identical in magnitude to our previous result. The fact that average station size in a county increases take place ahead of highway openings in the county does not appear to re‡ect the opening of highway sections outside of the county.

5.4 Do Our Results Di¤er With the Importance of Through Tra¢ c In the County?

Table 11 presents results where we interact (1 thrusharei) and thrusharei with xit. We …nd that the estimates on interactions between thrusharei and csmiit and csmiit disti are similar to what we show in Table 9. In contrast, 27 none of the none of the estimates on interactions between (1 thrusharei) and these variables are positive and signi…cant; indeed, the coe¢ cient on csmiit in the number of …rms regression is negative and signi…cant. As one might expect, our main results re‡ect changes in industry structure in counties where highway tra¢ c is large relative to local employment. There is no evidence that either the number or size of service stations in a county increased with the highway opened in the county in counties where highway tra¢ c is small relative to local employment, and there is some evidence that the number of stations decreased.

6 Conclusion

As described in the introduction, the opening of Interstate highway segments provides a fertile environment for studying how industries adjust to demand shocks. This paper presents evidence on the margin, timing, and magnitude of these adjustments in the case of service stations. We show how this industry, one in which a signi…cant share of capital investments are sunk to the industry and market, adjusts and how the adjustment process di¤ers depending on whether demand shifts spatially. Our empirical analysis shows the following. First, our sample period is one in which there is net exit in the aggregate which takes the form of an increase in the number of large stations and a decrease in the number of small ones. Second, we show that the increase in the number of large stations is associated with highway openings but the decrease in the number of small ones is not. The latter is consistent with the hypothesis that industry- and market-speci…c sunk investments make exit less sensitive to demand shocks than entry. Third, we show that the margin of adjustment systematically di¤ers, depending on whether the highway is located near to or far from the old route. When it is near, the adjustment takes the form of larger stations; in the long run, the increase in average station size is on the order of 6%. When it is far, it primarily takes the form of more stations; the estimated magnitude is on the order of 8%. The di¤erence in the margin of adjustment is what one would expect in light of monopolistic competition models, which imply that market size increases should lead to more …rms if price-cost margins are relatively insensitive to entry but larger …rms if they are sensitive to entry. Fourth, we show that the timing of adjustment systematically di¤ers along these lines as well. The adjustment begins two

28 years before the new highway opens when the new highway is close to the old route, and is complete the year the highway opens. In contrast, it begins the year the highway opens when the new highway is far from the old route. This di¤erence in timing is consistent with the hypothesis that, although preemption incentives might be greater when demand shocks create new market segments than when they do not, uncertainty is greater as well, and the latter leads …rms to delay investments in new capacity.

29 7 References

Baum-Snow, Nathaniel (2007). "Did Highways Cause Suburbanization?" Quarterly Journal of Economics, 122(2), 775-805.

Berry, Steven T. (1992). "Estimation of a Model of Entry in the Airline Industry." Econometrica, 60, 889-917.

Bresnahan, Timothy F. and Peter C. Reiss (1990). "Entry In Monopoly Markets." Review of Economic Studies, 57(3), 531-553.

Bresnahan, Timothy F. and Peter C. Reiss (1991). "Entry and Com- petition in Concentrated Markets." Journal of Political Economy, 99(5), 977-1009.

Bulow, Jeremy I., John D. Geanakoplos, and Paul D. Klemperer (1985). "Multimarket Oligopoly: Strategic Substitutes and Complements," Journal of Political Economy, 93(3), 488-511.

Campbell, Je¤rey and Hugo Hopenhayn (2005). "Market Size Matters." Journal of Industrial Economics, 53(1), 1-25.

Campbell, Je¤rey and Beverly Lapham (2004). "Real Exchange Rate Fluctuations and the Dynamics of Retail Trade Industries on the U.S.-Canada Border." American Economic Review, 94(4), 1194-1206

Chandra, Amitabh and Eric Thompson (2000). "Does Public Infrastruc- ture A¤ect Economic Activity? Evidence from the Interstate Highway Sys- tem." Regional Science and Urban Economics, 30(4), 457-490.

Dixit, Avinash K. and Robert S. Pindyck, Investment Under Uncertainty, Princeton University Press, 1994.

Dixit, Avinash K. and Joseph E. Stiglitz (1977). "Monopolistic Compe- tition and Optimum Product Diversity." American Economic Review, 67(3), 297-308.

Federal Highway Administration, Bene…ts of Interstate Highways, U.S. Department of Transportation, June 1970.

30 Fudenberg, Drew and Jean Tirole (1984). "The Fat Cat E¤ect, the Puppy Dog Ploy, and the Lean and Hungry Look," American Economic Review, 74(2), 361-368.

Fudenberg, Drew and Jean Tirole (1986). Dynamic Models of Oligopoly. Harwood, Chur, 1986.

Mazzeo, Michael J. (2002). "Product Choice and Oligopoly Market Structure." Rand Journal of Economics, 33(2), 221-242.

Michaels, Guy (2008). "The E¤ect of Trade on the Demand for Skill: Evidence from the Interstate Highway System." Review of Economics and Statistics, 90(4), 683-701.

Salop, Steven C. (1979). "Monopolistic Competition with Outside Goods." Bell Journal of Economics, 10(1), 141-156.

Spence, A. Michael (1976). "Product Selection, Fixed Costs, and Mo- nopolistic Competition." Review of Economic Studies, 43(2), 217-235.

Spence, A. Michael (1977). "Entry, Capacity, Investment and Oligopolis- tic Pricing." Bell Journal of Economics, 8(2), 534-544.

Spence, A. Michael (1979). "Investment Strategy and Growth in a New Market." Bell Journal of Economics, 10(1), 1-19.

Sutton, John (1991). Sunk Costs and Industry Structure. MIT Press, Cambridge, 1991.

Texas Transportation Institute (1966). Economic E¤ects of Bypasses and Freeways. Texas Transportation Institute, College Station, TX, 1966.

31 8 Appendix

8.1 De…nition of Corridors De…ning corridors …rst involves establishing the locations where corridors begin and end. These locations include most prominently major cities. After various trial de…nitions, we found that a useful and parsimonious way to generate a set of cities that serve as corridor endpoints is to look at the U.S. map in a standard road atlas. We found that cities listed in bold provided reasonable endpoints in the vast majority of cases. We spoke to cartographers at the …rm that produced the map, and asked the criteria for including a city in bold. We learned that all cities in bold have at least 100,000 population (for the map we use, in 1996), but not all cities that exceed this level are included on the map – suburban cities (e.g., Fullerton, CA) are excluded both because they are not major destinations and because including them would make the map cluttered. We asked the criteria for including these cities and were told “cartographic license.” In any case, this rule produces a very useful set of cities; central cities with at least 100,000 population. A list of these cities is in Table A1. In addition, we included the beginning and end of interstate highways as corridor endpoints, when the beginning or end of a highway (a) was not in a endpoint city, and (b) did not end at a junction with an interstate with the same orientation. One example of a corridor endpoint that satis…es this is Interstate 5’snorthern terminus at the Canadian border. Another is Interstate 4’s northern terminus at its intersection with Interstate 95; this is an intersection between a (even-numbered) east-west route and an (odd- numbered) north-south route. Within cities, we de…ned the beginning/end of the corridor to be at major intersections. The most common situation is where two interstates intersect near the heart of a city; when this occurs we use the interstate intersection as the placement for the node. (Sometimes the interstate intersection close to downtown is with a 3-digit highway.) In cities where there is a “dual-signed” segment where a single road is part of two two-digit interstate highways (e.g., Interstate 5-Interstate 10 in downtown Los Angeles), we use one of the endpoints of this dual-signed section. Where there is no interstate intersection near downtown, we use an important intersection close to downtown. This produces an easy division of some Interstate Highways into dis- tinct corridors. For example, it divides Interstate 25 into 4 corridors: start-

32 Albuquerque, Albuquerque-Colorado Springs, Colorado Springs-Denver, and Denver-end. This is simple because every mile of Interstate 25 belongs to only one corridor.

8.2 Highway Segments and Multiple Corridors Some segments of the interstate highway system belong to multiple corridors. The most common examples of this occur when an east-west interstate divides into two east-west interstates, and this division takes place outside one of our city endpoints: forks in the road. For example, Interstate 10 west of Tucson divides between Interstate 8, which goes to San Diego, and Interstate 10, which goes to Phoenix. The stretch of Interstate 10 that is west of Tuc- son but east of this fork is part of two corridors: Phoenix-Tucson and San Diego-Tucson. Another example of this is when highways merge then sepa- rate. For example, Interstate 70 and Interstate 76 come together southeast of Pittsburgh, continue together for a long stretch, then split. The “I70- I76” stretch is part of four corridors: Pittsburgh-Philadelphia, Pittsburgh- Baltimore, Columbus-Philadelphia, and Columbus-Baltimore. The adjacent segments are each part of two corridors; for example, the stretch of Inter- state 76 west of this dual signed stretch is part of Pittsburgh-Philadelphia and Pittsburgh-Baltimore. This pattern is common within metropolitan areas, albeit for much shorter stretches than I70-I76; as noted above, Interstate 5-Interstate 10 in Los Angeles is an example. A full list of highway stretches that are part of multiple corridors, and the corridors to which they are assigned, is available upon request from the authors.

8.3 Measuring Corridor Completion When Segments Are Part of Multiple Corridors An issue arises with respect to how to quantify how much of the corridor is complete in a county when highway segments are part of multiple corridors. For example, consider a county on Interstate 10 west of Tucson. This stretch of Interstate 10 is part of both Phoenix-Tucson and San Diego- Tucson. We construct our measure of corridor completion by …rst calculating the cumulative share of construction along each corridor, then weighting construction along the two corridors according to the tra¢ c volume on each of

33 the branches, measured at a point as close as possible to the “fork in the road;”in this case, tra¢ c volumes on Interstate 8 and Interstate 8 just west of where Interstate 10 splits into these two roads. We compute corridor-level construction variables analogously for all counties that are part of multiple corridors.

34 250000

200000

150000

100000

50000

0 1957 1962 1967 1972 1977 1982 1987 1992 1997

EC: Stations EC: Stations w/Payroll CBP: Reporting Units CBP: Stations w/Payroll

Figure 1. Service Stations in the United States. This Figure depicts Census counts of the number of service stations in the United States, and subsets thereof; these come from the Economic Census (EC) and County Business Patterns (CBP). The EC series show that the number of stations increased from the late 1950s to the early 1970s, then dropped sharply from then until the early 1980s. The CBP …gures report the number of …rms operating in each county before 1974, then the number of stations thereafter. The former falls relative to the EC-reported number of stations during the late 1960s and early 1970s, indicating that an increasing share of stations were owned by …rms that operated other stations in the same county.

35 9

7 6

3 2

0 1957 1962 1967 1972 1977 1982 1987 1992 1997 EC: Stations EC: Stations w/Payroll CBP: Reporting Units CBP: Stations w/Payroll

Figure 2. The Employment Size of Service Stations in the United States. This Figure depicts various measures of the employment size of service stations using data from the Economic Census (EC) and County Business Patterns (CBP). The EC series, which report employees per station using all stations and only stations with positive payroll, show that station size increased steadily throughout our sample period, increasing from 2.5 in 1964 to 5.6 in 1992. The CBP series report employees per "reporting unit" (…rm*county) before 1974, then employees per station thereafter. The former increases by more than employees per station during the late 1960s and early 1970s. Combined, the Figure indicates that stations’employment size roughly doubled between 1964-1992, and that about 1/4 of the increase in within-county …rm size between 1964-1973 is accounted for by an increase in the share of …rms that operated multiple stations in the same county.

36 S ervice S tation S ize, C haracteristics

G allons/ G allons/ Pumps/ Employees/ S hare S elf• S tation Pump S tation Pump S ervice S ales

1972 360.7 68.0 5.3 0.77 1977 508.8 97.4 5.2 0.88 30% 1982 543.1 90.2 6.0 0.86 63% 1987 697.4 97.1 7.2 0.85 75% 1992 802.8 110.8 7.2 0.88 91%

C hange 1972•1992 123% 63% 37% 15%

S hare of R evenues by Product C ategory

F uel, Tires, Food, Alcohol, O il P arts Tobacco Other

1972 82% 10% 2% 6% 1977 85% 5% 4% 6% 1982 88% 3% 5% 4% 1987 81% 2% 12% 6% 1992 79% 2% 15% 5%

S ource: Census of Retail Trade, Various Years.

Table 1. Service Station Size, Characteristics, and Revenue Sources. This Table reports how service stations’business and characteristics have changed between 1972-1992, using data from the Economic Census. Gallons per station more than doubled, re‡ecting increases in both gallons per pump and pumps per station. Employees per pump was constant starting in 1977. The self-service share of sales steadily increased to 91% by 1992. Automotive parts and accessories’share of station revenues decline between 1972-1982. The increase in convenience store-related sales increased sharply starting in 1982.

37 38

Figure 3. Sample Counties, Timing of Highway Completion, Distance from Old Route. T wo•Digit Highways In All T wo•Digit Highways Balanced Panel C ounties

C umulative C umulative Year Miles Share of Total Miles Share of Total

1960 7732 20% 3494 19% 1965 19423 50% 9273 49% 1970 29260 76% 14334 76% 1975 34884 90% 17138 91% 1980 37238 96% 18119 96% 1985 38065 98% 18571 99% 1990 38597 100% 18785 100%

1992 38665 100% 18833 100%

Table 2. Two-Digit Interstate Highway Completion. This Table depicts cumulative completed mileage of construction of "two-digit" Inter- state highways in all U.S. counties, and for the 677 counties in our balanced panel. Most of the mileage was completed during the 1960s and 1970s. The pace of highway completion in our balanced panel counties was similar to that overall.

39 Number of Firms/County

by Employment Size Category Total 1•3 4•7 8•19 20 or more

1964 45.8 35.7 7.6 2.2 0.4 1965 45.9 34.7 8.4 2.4 0.5 1966 45.8 33.4 9.1 2.8 0.5 1967 45.2 31.8 9.8 3.0 0.6 1968 45.2 30.3 10.7 3.5 0.7 1969 46.0 30.2 11.4 3.8 0.7 1970 45.3 29.2 11.5 3.9 0.7 1971 45.3 29.1 11.7 3.8 0.8 1972 45.7 27.9 12.6 4.3 0.8 1973 44.9 26.4 12.8 4.8 0.9

Number of S ervice S tations/County

by Employment Size Category Total 1•4 5•9 10•19 20 or more

1974 47.4 37.9 7.4 1.6 0.6 1975 44.9 33.3 9.0 2.0 0.6 1976 43.5 31.4 9.2 2.3 0.6 1977 43.3 30.8 9.7 2.2 0.6 1978 40.4 26.4 10.3 3.0 0.8 1979 37.6 23.7 10.2 2.8 0.9 1980 35.6 24.0 8.6 2.2 0.9 1981 33.8 22.2 8.7 2.2 0.7 1982 33.7 21.5 9.0 2.5 0.8 1983 35.9 23.0 9.7 2.5 0.7 1984 34.0 20.5 10.1 2.5 0.8 1985 32.1 18.3 9.9 3.0 1.0 1986 31.5 17.6 9.7 3.2 1.0 1987 33.6 18.2 10.6 3.6 1.1 1988 34.1 16.8 12.0 4.2 1.1 1989 33.4 15.8 11.8 4.5 1.3 1990 33.3 15.1 12.1 4.7 1.3 1991 32.4 14.9 11.7 4.5 1.4 1992 31.9 13.8 12.3 4.6 1.2

Table 3. Number of Firms and Service Stations per County, Overall and by Employment Size Category. This Table depicts the average number of …rms per county (in 1964-1973) and service stations per county (in 1974-1992) for counties in our balanced panel. Between 1964 and 1973, there is a decrease in the number of small …rms and an increase in the number of larger …rms. Between 1974 and 1992, the average number of service stations decreased by one-third, re‡ecting a large decrease in the number of small stations and a smaller increase in the number of large ones.

40 Date of Highway C ompletion L ate•E arly E arly Mid L a te R atio

1964 3.0 3.1 2.9 97% 1965 3.2 3.3 3.1 96% 1966 3.5 3.5 3.2 94% 1967 3.6 3.7 3.4 93% 1968 3.9 4.0 3.6 92% 1969 4.0 4.1 3.7 93% 1970 4.2 4.2 3.9 93% 1971 4.3 4.3 3.9 92% 1972 4.4 4.6 4.2 93% 1973 4.7 4.8 4.4 94% 1974 3.6 3.9 3.4 95% 1975 4.2 4.4 4.0 94% 1976 4.5 4.7 4.3 94% 1977 4.7 4.7 4.3 92% 1978 5.4 5.4 5.1 94% 1979 5.6 5.8 5.3 95% 1980 5.3 5.3 5.0 95% 1981 5.4 5.3 5.2 96% 1982 5.5 5.7 5.3 97% 1983 4.8 5.0 4.8 100% 1984 5.2 5.3 5.2 100% 1985 5.6 5.8 5.8 103% 1986 5.7 5.9 6.0 105% 1987 6.0 6.1 6.2 105% 1988 6.4 6.4 6.5 101% 1989 6.7 6.7 6.9 102% 1990 6.9 6.8 7.0 102% 1991 6.7 6.9 7.1 105% 1992 6.7 6.9 7.0 104%

Table 4. Average Employment Size of Firms/Stations, by Date of Highway Completion. This Table depicts average employment per …rm (in 1964-1973) and employment per station (in 1974-1992) for counties in our balanced panel with nonmissing employment for each sample year (N = 470). Highway completion is "early" if completed in 1965 or earlier, "mid" if completed between 1966-1971, and "late" if completed in or after 1972. N=167, 150, and 153 for early, mid, and late counties, respectively. The Table shows that while the employment size of …rms and stations increased in each of these categories, it took place earlier for the "early" counties than the "late" counties.

41 115%

110%

105%

100%

95%

90%

85% 1964 1968 1972 1976 1980 1984 1988 1992

Figure 4. Employment Size Ratios: "Late" Counties to "Early" Counties. This Table depicts the ratio of the average employment per …rm (in 1964-1973) and employment per station (in 1974-1992) for "late" and "early" counties in our balanced panel with nonmissing employment for each sample year (N = 470). Average …rm size increased in "early" counties relative to "late" counties early in our sample; average station size increased in "late" counties relative to "early" counties later in our sample.

42 120%

110%

100%

90%

80%

70%

60% 1964 1968 1972 1976 1980 1984 1988 1992

s mall medium large

Figure 5. Firm and Station Count Ratios: "Late" Counties to "Early" Counties. This Table depicts the ratio of the average number of …rms (in 1964-1973) or stations (in 1974-1992), by size category, in "late" and "early" counties in our balanced panel (N = 677). In 1964-1973, "small," "medium," and "large" include …rms with 1-3, 4-7, and 8-19 employees, respectively. In 1974-1992, "small," "medium," and "large" include stations with 1-4, 5-9, and 10-19 employees, respectively. This Table shows increases in the number of large, and to some extent medium-sized, …rms and stations took place earlier in the "early" counties than the "late" counties. It shows no evidence that changes in the number of small …rms and stations were related to highway completion dates.

43 Table 5 VARs of the Number and Average E mployment S ize of S ervice S tations on Highway Openings

Autoregressive Coefficients Cumulative S hare of Highway Opened in County S um of Distributed Lag: Years From Highway Opening Lag Coefficients ln(nit•1) ln(ait•1) •3 •2 •1 0 1 2 3

No Leads or Lags

ln(nit) 0.768 0.029 0.002 0.002 (0.005) (0.004) (0.005) (0.005)

ln(ait) 0.032 0.638 0.019 0.019 (0.007) (0.006) (0.007) (0.007)

One Lead and Lag

ln(nit) 0.766 0.032 •0.006 •0.003 0.011 0.003 (0.005) (0.004) (0.011) (0.014) (0.010) (0.006)

ln(ait) •0.035 0.635 0.030 0.002 •0.010 0.022 (0.007) (0.006) (0.015) (0.019) (0.013) (0.008) 44 Two Leads and Lags

ln(nit) 0.754 0.033 •0.010 0.003 •0.001 0.009 0.002 0.002 (0.005) (0.005) (0.013) (0.016) (0.015) (0.014) (0.010) (0.007)

ln(ait) 0.040 0.618 0.011 0.018 0.007 •0.036 0.025 0.025 (0.008) (0.007) (0.018) (0.022) (0.021) (0.019) (0.013) (0.009)

Three Leads and Lags

ln(nit) 0.737 0.032 •0.022 0.019 •0.013 0.000 0.015 0.007 •0.012 •0.007 (0.006) (0.005) (0.015) (0.019) (0.018) (0.016) (0.015) (0.014) (0.009) (0.008)

ln(ait) 0.039 0.598 •0.001 0.013 0.016 0.013 •0.055 0.032 0.006 0.024 (0.008) (0.007) (0.021) (0.026) (0.025) (0.023) (0.021) (0.020) (0.013) (0.012)

These results are from county•level VAR specifications that relate the number of service stations (per 1980 employment) and the average employment size of service stations to Interstate highway openings. The specifications also include county and year fixed effects (not reported). We also allow the autoregressive coefficients to differ in year 1974 to accommodate Census' change in reporting units between 1973 and 1974.

These results use all counties with non•missing reports for number of firms/establishments from 1964•1992, N=677. S tandard errors are in parentheses; bold indicates that the estimates is statistically significantly different zero using a test of size 0.05. 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 •0.01 •3 •2 •1 0 1 2 3 4 5

None One Two Three 45 Figure 6. Impulse-Response Functions for Highway Openings on Employment Size of Service Stations, For Speci…cations with Zero to Three Leads and Lags. This graph depicts how the average size of service stations change around the time that Interstate highway segments are completed in a county, for speci…cations that use zero to three leads and lags. The vertical axes scaled in log-points; 0.04 represents a 4% increase. The horizontal axis is years from segment completion; "-2" means two years before a segment is completed. These graphs illustrate that average station size began to increase during the two years preceding the new highway’scompletion. All speci…cations capture a similar long-run change in average station size. Table 6 VARs of the Number of Service Stations in Employment Size Categories on Highway Openings

Autoregressive C oefficients C umulatiive S hare of Highway Opened in C ounty S um of Distributed L ag: Y ears from Highway O pening L ag C oefficients s1 s2 s3 s4 •2 •1 0 1 2

No Leads or Lags s1 0.802 0.149 •0.017 •0.063 0.179 0.179 (0.004) (0.007) (0.015) (0.036) (0.156) (0.156) s2 0.077 0.650 0.271 0.114 0.080 0.080 (0.003) (0.006) (0.011) (0.026) (0.110) (0.110) s3 •0.005 0.082 0.579 0.325 0.234 0.234 (0.002) (0.003) (0.006) (0.015) (0.064) (0.064) s4 •0.002 •0.001 0.063 0.519 0.010 0.010 (0.001) (0.001) (0.003) (0.006) (0.028) (0.028)

One Lead and Lag s1 0.794 0.158 •0.001 •0.021 0.214 •0.401 •0.450 0.263 (0.004) (0.008) (0.016) (0.038) (0.333) (0.429) (0.295) (0.172) s2 0.079 0.640 0.268 0.144 0.008 •0.088 0.193 0.113 (0.003) (0.006) (0.011) (0.027) (0.236) (0.303) (0.208) (0.122) s3 •0.004 0.086 0.573 0.293 0.304 •0.052 0.038 0.290 (0.002) (0.003) (0.006) (0.016) (0.136) (0.175) (0.121) (0.070) s4 •0.004 •0.003 0.064 0.545 0.020 •0.031 0.031 0.020 (0.001) (0.001) (0.003) (0.007) (0.058) (0.074) (0.051) (0.030)

Two Leads and Lags s1 0.778 0.172 0.010 •0.007 0.035 0.336 •0.428 0.273 0.017 0.232 (0.004) (0.009) (0.017) (0.040) (0.384) (0.489) (0.459) (0.431) (0.298) (0.206) s2 0.082 0.622 0.277 0.179 •0.058 •0.143 •0.071 0.358 0.031 0.117 (0.003) (0.006) (0.012) (0.029) (0.271) (0.345) (0.325) (0.305) (0.210) (0.145) s3 •0.006 0.083 0.569 0.303 0.408 0.022 •0.016 •0.183 0.226 0.457 (0.002) (0.004) (0.007) (0.016) (0.156) (0.198) (0.186) (0.175) (0.121) (0.083) s4 •0.003 0.002 0.059 0.525 •0.065 0.104 •0.031 •0.040 0.052 0.021 (0.001) (0.002) (0.003) (0.007) (0.066) (0.084) (0.079) (0.074) (0.051) (0.035)

T hese results are from county•level VAR specifications that relate the number of service stations in different size categories to the share of interstate highway mileage in the county that had opened by year t. S 1, S 2, S 3, and S 4 consist of firms with 1•3, 4•7, 8•19, and 20 or more employees in the county (these categories are 1•4, 5•9, 10•19, and 20 or more after 1974). The specifications also include county and year fixed effects (not reported). We also allow the autoregressive coefficients to differ in year 1974 to accommodate Census' change in reporting units between 1973 and 1974.

46 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 •3 •2 •1 0 1 2 3 4 5

None One Two Three

47 Figure 7. Impulse-Response Functions for Highway Openings on Number of "Large" Service Stations, For Speci…cations with Zero to Three Leads and Lags. This graph depicts how the number of "large" service stations change around the time that Interstate highway segments are completed in a county, for speci…cations that use zero to three leads and lags. The vertical axis is the number of stations (before 1974, …rms). The horizontal axis is years from segment completion; "-2" means two years before a segment is completed. These graphs illustrate that the number of large service stations began to increase during the two years preceding the new highway’scompletion. Table 7 VARs of the Number and Average Employment Size of Service Stations on Highway Openings Interactions with Post 73 Dummy

Autoregressive C oefficients C umulative S hare of Highway Opened in C ounty Post 73*Cumulative S hare of Highway Opened in C ounty Distributed Lag: Y ears from Highway Opening Distributed Lag: Y ears from Highway Opening ln(nit•1) ln(ait•1) •2 •1 0 1 2 •2 •1 0 1 2

No Leads or Lags

ln(nit) 0.768 0.029 •0.001 0.013 (0.005) (0.004) (0.005) (0.011)

ln(ait) 0.032 0.638 0.020 •0.007 (0.007) (0.006) (0.007) (0.016)

One Lead and Lag 48 ln(nit) 0.766 0.032 •0.006 •0.003 0.009 0.009 0.002 0.008 (0.005) (0.004) (0.012) (0.015) (0.010) (0.017) (0.021) (0.017)

ln(ait) 0.039 0.617 0.035 0.002 •0.013 •0.024 0.002 0.020 (0.008) (0.007) (0.016) (0.020) (0.014) (0.023) (0.029) (0.023)

Two Leads and Lags

ln(nit) 0.754 0.033 0.005 •0.008 •0.001 0.008 0.000 •0.043 0.040 0.001 0.000 0.010 (0.005) (0.005) (0.014) (0.017) (0.016) (0.015) (0.010) (0.017) (0.022) (0.021) (0.021) (0.017)

ln(ait) 0.039 0.618 0.004 0.029 0.008 •0.032 0.017 0.025 •0.038 •0.003 •0.009 0.044 (0.008) (0.006) (0.019) (0.024) (0.022) (0.021) (0.014) (0.024) (0.030) (0.030) (0.029) (0.025)

These results are from county•level VAR specifications that relate the number of service stations (per 1980 employment) and the average employment size of service stations to Interstate highway openings. The specifications also include county and year fixed effects (not reported). We also allow the autoregressive coefficients to differ in year 1974 to accommodate C ensus' change in reporting units between 1973 and 1974.

These results use all counties with non•missing reports for number of firms/establishments from 1964•1992, N=677. S tandard errors are in parentheses; bold indicates that the estimates is statistically significantly different zero using a test of size 0.05. Table 8 Interaction S pecifications with Distance From Old Route: No L eads or Lags

Dependent Variable: ln(nit)

C umulative S hare of Highway Opened ("C S HO") 0.002 •0.006 •0.013 (0.005) (0.006) (0.008)

C S HO*distance from old route 0.0027 0.0076 (0.0012) (0.0035)

C S HO*(distance from old route)^2 •0.0004 (0.0002)

Dependent Variable: ln(ait)

C umulative S hare of Highway Opened ("C S HO") 0.019 0.019 0.016 (0.007) (0.008) (0.011) 49 C S HO*distance from old route •0.0002 0.0022 (0.0017) (0.0049)

C S HO*(distance from old route)^2 •0.0002 (0.0003)

These results are from county•level interactions that relate the number of service stations (per 1980 employment) and the average employment size of service stations to interstate highway openings.

The first column is the same as the top panel of Table X. The second and third interact the cumulative share of highway opened in the county (C S HO) with the average distance between the interstate and the old route that it replaced in the county.

The results indicate that highway openings have a stronger effect on the number of firms when the old route is farther from the interstate, and that distance has a diminishing incremental effect. In contrast, there is no evidence the relationship between average service station size and highway openings varies with distance from old route. Table 9 VARs of the Number and Average Employment S ize of S ervice S tations on Highway Openings Distance Interactions, Two Leads and Lags

Autoregressive Coefficients C umulative S hare of Highway Opened in C ounty Distance*Cumulative S hare of Highway Opened in C ounty S um of Distributed Lag: Y ears from Highway Opening Distributed Lag: Y ears from Highway Opening Lag Coefficients ln(nit•1) ln(ait•1) •2 •1 0 1 2 •2 •1 0 1 2 d=0 d=10

No Leads or Lags

ln(nit) 0.768 0.030 •0.006 0.003 •0.005 0.021 (0.005) (0.004) (0.006) (0.001) (0.006) (0.010)

ln(ait) 0.039 0.617 0.020 0.000 0.020 0.018 (0.008) (0.007) (0.009) (0.002) (0.009) (0.014)

One Lead and Lag 50 ln(nit) 0.765 0.032 •0.010 0.007 •0.005 0.001 •0.003 0.005 •0.007 0.024 (0.005) (0.004) (0.015) (0.018) (0.013) (0.003) (0.005) (0.002) (0.007) (0.011)

ln(ait) 0.039 0.617 0.054 •0.016 •0.011 •0.008 0.006 0.000 0.028 0.010 (0.008) (0.007) (0.020) (0.026) (0.017) (0.005) (0.005) (0.003) (0.010) (0.015)

Two Leads and Lags

ln(nit) 0.754 0.033 •0.010 •0.010 0.016 •0.010 •0.003 0.000 0.001 •0.005 0.006 0.001 •0.008 0.024 (0.005) (0.005) (0.017) (0.021) (0.020) (0.188) (0.013) (0.004) (0.005) (0.004) (0.004) (0.003) (0.009) (0.013)

ln(ait) 0.039 0.617 0.025 0.038 •0.011 •0.028 0.010 •0.005 •0.007 0.006 •0.002 0.005 0.035 0.005 (0.008) (0.007) (0.024) (0.030) (0.028) (0.026) (0.018) (0.005) (0.006) (0.006) (0.005) (0.003) (0.012) (0.018)

These results are from county•level VAR specifications that relate the number of service stations (per 1980 employment) and the average employment size of service stations to Interstate highway openings. The specifications also include county and year fixed effects (not reported). We also allow the autoregressive coefficients to differ in year 1974 to accommodate C ensus' change in reporting units between 1973 and 1974.

These results use all counties with non•missing reports for number of firms/establishments from 1964•1992, N=677. S tandard errors are in parentheses; bold indicates that the estimates is statistically significantly different zero using a test of size 0.05. Number of Service Stations and Year From Interstate Highway Opening

0.1

0.08

0.06

0.04

0.02

0 •2 •1 0 1 2 3 4 5 6 •0.02

•0.04

0 miles 1.25 miles 10 miles

Average Service Station Employment Size and Year From Interstate Highway Opening

0.1

0.08

0.06

0.04

0.02

0 •2 •1 0 1 2 3 4 5 6 •0.02

•0.04

0 miles 1.25 miles 10 miles

Figure 8. Impulse-Response Functions for Highway Openings on Market Structure of Service Stations, by Distance from Old Route. These graphs depict how the number and average size of service stations change around the time that Interstate highway segments are completed in a county, and how this di¤ers with how close the Interstate is from the previous route. The vertical axes scaled in log-points; 0.04 represents a 4% increase. The horizontal axis is years from segment completion; "-2" means two years before a segment is completed. These graphs illustrate that

51 when the Interstate was close to the old route, the industry adjustment was in an increase in average station size during the two years preceding the new highway’scompletion. When it was far, the adjustment was an increase in the number of stations that took place after the new highway was completed.

52 Table 10 VARs of the Number and Average Employment Size of Service Stations on Highway Openings Distance Interactions, One Lead and Lag

Autoregressive Coefficients Cumulative Share of Highway Cumulative Share of Highway Cumulative Share of Highway Sum of Opened in Corridor Opened in County Opened in County*Distance Lag Coefficients ln(nit•1) ln(ait•1) •1 0 1 •1 0 1 •1 0 1 d=0 d=10

No Leads or Lags

ln(nit) 0.768 0.030 0.005 •0.008 0.003 0.005 0.019 (0.005) (0.004) (0.010) (0.007) (0.001) (0.010) (0.010)

ln(ait) 0.031 0.637 0.014 0.016 0.000 0.016 0.012 (0.007) (0.006) (0.014) (0.010) (0.002) (0.010) (0.015) 53

One Lead and Lag

ln(nit) 0.765 0.032 •0.025 0.009 0.011 •0.006 0.009 •0.008 0.001 •0.002 0.005 •0.004 0.026 (0.005) (0.004) (0.019) (0.022) (0.016) (0.015) (0.019) (0.013) (0.003) (0.004) (0.002) (0.008) (0.012)

ln(ait) 0.034 0.634 0.001 0.009 0.009 0.052 •0.013 •0.019 •0.008 0.006 0.000 0.021 0.001 (0.007) (0.006) (0.027) (0.031) (0.023) (0.020) (0.026) (0.018) (0.005) (0.006) (0.003) (0.012) (0.016)

These results use all counties with non•missing reports for number of firms/establishments from 1964•1992, N=677. Standard errors are in parentheses; bold indicates that the estimates is statistically significantly different zero using a test of size 0.05. Table 11 VARs of the Number and Average Employment Size of Service Stations on Highway Openings Distance Interactions, One Lead and Lag

Autoregressive Coefficients Cumulative Share of Highway Cumulative Share of Highway Sum of Opened in County Opened in County*Distance Lag Coefficients ln(nit•1) ln(ait•1) •1 0 1 •1 0 1 d=0 d=10

No Leads or Lags Interactions with "thrushare" ln(nit) 0.766 0.029 0.017 0.005 0.017 0.071 (0.005) (0.004) (0.012) (0.003) (0.012) (0.020) ln(ait) 0.030 0.636 0.067 •0.005 0.067 0.019 (0.007) (0.006) (0.017) (0.004) (0.017) (0.028)

Interactions with (1•"thrushare") ln(nit) •0.034 •0.001 •0.034 •0.044 (0.015) (0.003) (0.015) (0.026) ln(ait) •0.037 0.006 •0.037 0.019 (0.020) (0.005) (0.020) (0.035)

One Lead and Lag Interactions with "thrushare" ln(nit) 0.764 0.032 0.011 •0.020 0.029 0.003 •0.001 0.004 0.019 0.079 (0.005) (0.004) (0.030) (0.039) (0.026) (0.007) (0.008) (0.005) (0.014) (0.022) ln(ait) 0.032 0.633 0.077 •0.023 0.024 •0.023 0.012 0.002 0.077 •0.016 (0.007) (0.006) (0.042) (0.054) 0.036 (0.009) (0.010) (0.007) (0.018) (0.031)

Interactions with (1•"thrushare") ln(nit) •0.030 0.038 •0.046 •0.003 •0.004 0.005 •0.038 •0.049 (0.038) (0.049) (0.033) (0.009) (0.011) (0.007) (0.016) (0.028) ln(ait) 0.018 0.004 •0.057 0.014 •0.003 •0.003 •0.036 0.044 (0.053) (0.068) (0.046) (0.012) (0.015) (0.011) (0.023) (0.039)

54 Table A1 Cities that Are Corridor Endpoints

Abilene, TX Detroit, MI Madison, WI San Bernardino, CA Akron, OH Durham, NC Memphis, TN San Diego, CA Albany, NY El Paso, TX Miami, FL San Francisco, CA Albuquerque, NM Erie, PA Milwaukee, WI Savannah, GA Allentown, PA Eugene, OR Minneapolis, MN Seattle, WA Amarillo, TX Flint, MI Mobile, AL Shreveport, LA Ann Arbor, MI Fort Lauderdale, FL Montgomery, AL Sioux Falls, SD Atlanta, GA Fort Wayne, IN Nashville, TN South Bend, IN Austin, TX Fort Worth, TX New Haven, CT Spokane, WA Baltimore, MD Gary, IN New Orleans, LA Springfield, IL Baton Rouge, LA Grand Rapids, MI New York, NY Springfield, MO Beaumont, TX Greensboro, NC Newark, NJ St. Louis, MO Birmingham, AL Hartford, CT Norfolk, VA Stockton, CA Boise, ID Houston, TX Oklahoma City, OK Syracuse, NY Boston, MA Indianapolis, IN Omaha, NE Tacoma, WA Bridgeport, CT Jackson, MS Orlando, FL Tallahassee, FL Buffalo, NY Jacksonville, FL Peoria, IL Tampa, FL Charlotte, NC Kansas City, MO Philadelphia, PA Toledo, OH Chattanooga, TN Knoxville, TN Phoenix, AZ Topeka, KS Chicago, IL Lafayette, LA Pittsburgh, PA Tucson, AZ Cincinnati, OH Lansing, MI Portland, OR Tulsa, OK Cleveland, OH Laredo, TX Providence, RI Waco, TX Colorado Springs, CO Las Vegas, NV Raleigh, NC Washington, DC Columbia, SC Lexington, KY Reno, NV Wichita, KS Columbus, OH Lincoln, NE Richmond, VA Winston Salem, NC Corpus Christi, TX Little Rock, AR Rockford, IL Dallas, TX Los Angeles, CA Sacramento, CA Dayton, OH Louisville, KY Salem, OR Denver, CO Lubbock, TX Salt Lake City, UT Des Moines, IA Macon, GA San Antonio, TX

55 Working Paper Series

A series of research studies on regional economic issues relating to the Seventh Federal Reserve District, and on financial and economic topics.

U.S. Corporate and Bank Insolvency Regimes: An Economic Comparison and Evaluation WP-06-01 Robert R. Bliss and George G. Kaufman

Redistribution, Taxes, and the Median Voter WP-06-02 Marco Bassetto and Jess Benhabib

Identification of Search Models with Initial Condition Problems WP-06-03 Gadi Barlevy and H. N. Nagaraja

Tax Riots WP-06-04 Marco Bassetto and Christopher Phelan

The Tradeoff between Mortgage Prepayments and Tax-Deferred Retirement Savings WP-06-05 Gene Amromin, Jennifer Huang,and Clemens Sialm

Why are safeguards needed in a trade agreement? WP-06-06 Meredith A. Crowley

Taxation, Entrepreneurship, and Wealth WP-06-07 Marco Cagetti and Mariacristina De Nardi

A New Social Compact: How University Engagement Can Fuel Innovation WP-06-08 Laura Melle, Larry Isaak, and Richard Mattoon

Mergers and Risk WP-06-09 Craig H. Furfine and Richard J. Rosen

Two Flaws in Business Cycle Accounting WP-06-10 Lawrence J. Christiano and Joshua M. Davis

Do Consumers Choose the Right Credit Contracts? WP-06-11 Sumit Agarwal, Souphala Chomsisengphet, Chunlin Liu, and Nicholas S. Souleles

Chronicles of a Deflation Unforetold WP-06-12 François R. Velde

Female Offenders Use of Social Welfare Programs Before and After Jail and Prison: Does Prison Cause Welfare Dependency? WP-06-13 Kristin F. Butcher and Robert J. LaLonde

Eat or Be Eaten: A Theory of Mergers and Firm Size WP-06-14 Gary Gorton, Matthias Kahl, and Richard Rosen

1 Working Paper Series (continued)

Do Bonds Span Volatility Risk in the U.S. Treasury Market? A Specification Test for Affine Term Structure Models WP-06-15 Torben G. Andersen and Luca Benzoni

Transforming Payment Choices by Doubling Fees on the Illinois Tollway WP-06-16 Gene Amromin, Carrie Jankowski, and Richard D. Porter

How Did the 2003 Dividend Tax Cut Affect Stock Prices? WP-06-17 Gene Amromin, Paul Harrison, and Steven Sharpe

Will Writing and Bequest Motives: Early 20th Century Irish Evidence WP-06-18 Leslie McGranahan

How Professional Forecasters View Shocks to GDP WP-06-19 Spencer D. Krane

Evolving Agglomeration in the U.S. auto supplier industry WP-06-20 Thomas Klier and Daniel P. McMillen

Mortality, Mass-Layoffs, and Career Outcomes: An Analysis using Administrative Data WP-06-21 Daniel Sullivan and Till von Wachter

The Agreement on Subsidies and Countervailing Measures: Tying One’s Hand through the WTO. WP-06-22 Meredith A. Crowley

How Did Schooling Laws Improve Long-Term Health and Lower Mortality? WP-06-23 Bhashkar Mazumder

Manufacturing Plants’ Use of Temporary Workers: An Analysis Using Census Micro Data WP-06-24 Yukako Ono and Daniel Sullivan

What Can We Learn about Financial Access from U.S. Immigrants? WP-06-25 Una Okonkwo Osili and Anna Paulson

Bank Imputed Interest Rates: Unbiased Estimates of Offered Rates? WP-06-26 Evren Ors and Tara Rice

Welfare Implications of the Transition to High Household Debt WP-06-27 Jeffrey R. Campbell and Zvi Hercowitz

Last-In First-Out Oligopoly Dynamics WP-06-28 Jaap H. Abbring and Jeffrey R. Campbell

Oligopoly Dynamics with Barriers to Entry WP-06-29 Jaap H. Abbring and Jeffrey R. Campbell

Risk Taking and the Quality of Informal Insurance: Gambling and Remittances in Thailand WP-07-01 Douglas L. Miller and Anna L. Paulson

2 Working Paper Series (continued)

Fast Micro and Slow Macro: Can Aggregation Explain the Persistence of Inflation? WP-07-02 Filippo Altissimo, Benoît Mojon, and Paolo Zaffaroni

Assessing a Decade of Interstate Bank Branching WP-07-03 Christian Johnson and Tara Rice

Debit Card and Cash Usage: A Cross-Country Analysis WP-07-04 Gene Amromin and Sujit Chakravorti

The Age of Reason: Financial Decisions Over the Lifecycle WP-07-05 Sumit Agarwal, John C. Driscoll, Xavier Gabaix, and David Laibson

Information Acquisition in Financial Markets: a Correction WP-07-06 Gadi Barlevy and Pietro Veronesi

Monetary Policy, Output Composition and the Great Moderation WP-07-07 Benoît Mojon

Estate Taxation, Entrepreneurship, and Wealth WP-07-08 Marco Cagetti and Mariacristina De Nardi

Conflict of Interest and Certification in the U.S. IPO Market WP-07-09 Luca Benzoni and Carola Schenone

The Reaction of Consumer Spending and Debt to Tax Rebates – Evidence from Consumer Credit Data WP-07-10 Sumit Agarwal, Chunlin Liu, and Nicholas S. Souleles

Portfolio Choice over the Life-Cycle when the Stock and Labor Markets are Cointegrated WP-07-11 Luca Benzoni, Pierre Collin-Dufresne, and Robert S. Goldstein

Nonparametric Analysis of Intergenerational Income Mobility WP-07-12 with Application to the United States Debopam Bhattacharya and Bhashkar Mazumder

How the Credit Channel Works: Differentiating the Bank Lending Channel WP-07-13 and the Balance Sheet Channel Lamont K. Black and Richard J. Rosen

Labor Market Transitions and Self-Employment WP-07-14 Ellen R. Rissman

First-Time Home Buyers and Residential Investment Volatility WP-07-15 Jonas D.M. Fisher and Martin Gervais

Establishments Dynamics and Matching Frictions in Classical Competitive Equilibrium WP-07-16 Marcelo Veracierto

Technology’s Edge: The Educational Benefits of Computer-Aided Instruction WP-07-17 Lisa Barrow, Lisa Markman, and Cecilia Elena Rouse

3 Working Paper Series (continued)

The Widow’s Offering: Inheritance, Family Structure, and the Charitable Gifts of Women WP-07-18 Leslie McGranahan

Demand Volatility and the Lag between the Growth of Temporary and Permanent Employment WP-07-19 Sainan Jin, Yukako Ono, and Qinghua Zhang

A Conversation with 590 Nascent Entrepreneurs WP-07-20 Jeffrey R. Campbell and Mariacristina De Nardi

Cyclical Dumping and US Antidumping Protection: 1980-2001 WP-07-21 Meredith A. Crowley

Health Capital and the Prenatal Environment: The Effect of Maternal Fasting During Pregnancy WP-07-22 Douglas Almond and Bhashkar Mazumder

The Spending and Debt Response to Minimum Wage Hikes WP-07-23 Daniel Aaronson, Sumit Agarwal, and Eric French

The Impact of Mexican Immigrants on U.S. Wage Structure WP-07-24 Maude Toussaint-Comeau

A Leverage-based Model of Speculative Bubbles WP-08-01 Gadi Barlevy

Displacement, Asymmetric Information and Heterogeneous Human Capital WP-08-02 Luojia Hu and Christopher Taber

BankCaR (Bank Capital-at-Risk): A credit risk model for US commercial bank charge-offs WP-08-03 Jon Frye and Eduard Pelz

Bank Lending, Financing Constraints and SME Investment WP-08-04 Santiago Carbó-Valverde, Francisco Rodríguez-Fernández, and Gregory F. Udell

Global Inflation WP-08-05 Matteo Ciccarelli and Benoît Mojon

Scale and the Origins of Structural Change WP-08-06 Francisco J. Buera and Joseph P. Kaboski

Inventories, Lumpy Trade, and Large Devaluations WP-08-07 George Alessandria, Joseph P. Kaboski, and Virgiliu Midrigan

School Vouchers and Student Achievement: Recent Evidence, Remaining Questions WP-08-08 Cecilia Elena Rouse and Lisa Barrow

4 Working Paper Series (continued)

Does It Pay to Read Your Junk Mail? Evidence of the Effect of Advertising on Home Equity Credit Choices WP-08-09 Sumit Agarwal and Brent W. Ambrose

The Choice between Arm’s-Length and Relationship Debt: Evidence from eLoans WP-08-10 Sumit Agarwal and Robert Hauswald

Consumer Choice and Merchant Acceptance of Payment Media WP-08-11 Wilko Bolt and Sujit Chakravorti

Investment Shocks and Business Cycles WP-08-12 Alejandro Justiniano, Giorgio E. Primiceri, and Andrea Tambalotti

New Vehicle Characteristics and the Cost of the Corporate Average Fuel Economy Standard WP-08-13 Thomas Klier and Joshua Linn

Realized Volatility WP-08-14 Torben G. Andersen and Luca Benzoni

Revenue Bubbles and Structural Deficits: What’s a state to do? WP-08-15 Richard Mattoon and Leslie McGranahan

The role of lenders in the home price boom WP-08-16 Richard J. Rosen

Bank Crises and Investor Confidence WP-08-17 Una Okonkwo Osili and Anna Paulson

Life Expectancy and Old Age Savings WP-08-18 Mariacristina De Nardi, Eric French, and John Bailey Jones

Remittance Behavior among New U.S. Immigrants WP-08-19 Katherine Meckel

Birth Cohort and the Black-White Achievement Gap: The Roles of Access and Health Soon After Birth WP-08-20 Kenneth Y. Chay, Jonathan Guryan, and Bhashkar Mazumder

Public Investment and Budget Rules for State vs. Local Governments WP-08-21 Marco Bassetto

Why Has Home Ownership Fallen Among the Young? WP-09-01 Jonas D.M. Fisher and Martin Gervais

Why do the Elderly Save? The Role of Medical Expenses WP-09-02 Mariacristina De Nardi, Eric French, and John Bailey Jones

Using Stock Returns to Identify Government Spending Shocks WP-09-03 Jonas D.M. Fisher and Ryan Peters

5 Working Paper Series (continued)

Stochastic Volatility WP-09-04 Torben G. Andersen and Luca Benzoni

The Effect of Disability Insurance Receipt on Labor Supply WP-09-05 Eric French and Jae Song

CEO Overconfidence and Dividend Policy WP-09-06 Sanjay Deshmukh, Anand M. Goel, and Keith M. Howe

Do Financial Counseling Mandates Improve Mortgage Choice and Performance? WP-09-07 Evidence from a Legislative Experiment Sumit Agarwal,Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet, and Douglas D. Evanoff

Perverse Incentives at the Banks? Evidence from a Natural Experiment WP-09-08 Sumit Agarwal and Faye H. Wang

Pay for Percentile WP-09-09 Gadi Barlevy and Derek Neal

The Life and Times of Nicolas Dutot WP-09-10 François R. Velde

Regulating Two-Sided Markets: An Empirical Investigation WP-09-11 Santiago Carbó Valverde, Sujit Chakravorti, and Francisco Rodriguez Fernandez

The Case of the Undying Debt WP-09-12 François R. Velde

Paying for Performance: The Education Impacts of a Community College Scholarship Program for Low-income Adults WP-09-13 Lisa Barrow, Lashawn Richburg-Hayes, Cecilia Elena Rouse, and Thomas Brock

Establishments Dynamics, Vacancies and Unemployment: A Neoclassical Synthesis WP-09-14 Marcelo Veracierto

The Price of Gasoline and the Demand for Fuel Economy: Evidence from Monthly New Vehicles Sales Data WP-09-15 Thomas Klier and Joshua Linn

Estimation of a Transformation Model with Truncation, Interval Observation and Time-Varying Covariates WP-09-16 Bo E. Honoré and Luojia Hu

Self-Enforcing Trade Agreements: Evidence from Antidumping Policy WP-09-17 Chad P. Bown and Meredith A. Crowley

Too much right can make a wrong: Setting the stage for the financial crisis WP-09-18 Richard J. Rosen

Can Structural Small Open Economy Models Account for the Influence of Foreign Disturbances? WP-09-19 Alejandro Justiniano and Bruce Preston 6 Working Paper Series (continued)

Liquidity Constraints of the Middle Class WP-09-20 Jeffrey R. Campbell and Zvi Hercowitz

Monetary Policy and Uncertainty in an Empirical Small Open Economy Model WP-09-21 Alejandro Justiniano and Bruce Preston

Firm boundaries and buyer-supplier match in market transaction: IT system procurement of U.S. credit unions WP-09-22 Yukako Ono and Junichi Suzuki

Health and the Savings of Insured Versus Uninsured, Working-Age Households in the U.S. WP-09-23 Maude Toussaint-Comeau and Jonathan Hartley

The Economics of “Radiator Springs:” Industry Dynamics, Sunk Costs, and Spatial Demand Shifts WP-09-24 Jeffrey R. Campbell and Thomas N. Hubbard